H.K. Moffatt - Generation of Magnetic Fields by Fluid Motion

www.moffatt.
tc
ADVANCES
Reprinted trom:
I N
APPLIED
@ 19/6
MECHANICS. VOL 16
bndm
ACADEMIC P&=. New York San Francisco
Generation of Magnetic Fields by Fluid Motion

H. K . MOFFATT
Department o Applied Mathematics and Theoretical Physics f f University o Cambridge Cambridge. England
. .
.. I
I . Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 120 I1. Magnetokinematic Preliminaries . . . . . . . . . . . . . . . . . . 125 A . Idealization of the Kinematic Dynamo Problem . . . . . . . . . . 125 B. Magnetic Field Representations . . . . . . . . . . . . . . . . . 127 C. Alfven's Theorem and Woltjer's Invariant . . . . . . . . . . . . . 128 D. Natural Decay Modes and Force-Free Fields . . . . . . . . . . . 129 1 1 Convection, Distortion. and Diffusion of B Lines . . . . . . . . . . . 130 1. A . Balance of Stretching and Diffusion in a Magnetic Flux Rope . . . . 131 B. Flux Expulsion by Flows with Closed Streamlines . . . . . . . . . 131 C. Topological Pumping of Magnetic Flux . . . . . . . . . . . . . . 133 D. Generation of Toroidal Field by Differential Rotation . . . . . . . 134 IV. Some Basic Results . . . . . . . . . . . . . . . . . . . . . . . . 135 A . Cowling's Theorem and Related Results . . . . . . . . . . . . . 135 B. Rotor Dynamos . . . . . . . . . . . . . . . . . . . . . . . . 137 V . The Mean Electromotive Force Generated by a Random Velocity Field . 139 A . The Two-Scale Approach . . . . . . . . . . . . . . . . . . . . 139 B The Strong Diffusion Limit . . . . . . . . . . . . . . . . . . . 141 C. Evaluation of aij and But for a Random Wave Field . . . . . . . . 145 D. Effect of Turbulence in the Weak Diffusion Limit, 1 4 0 . . . . . . 147 E. The Forms of ay and But in Axisymmetric Turbulence . . . . . . . 150 F . Dynamo Equations for Axisymmetric Mean Fields Including Mean Flow Effects . . . . . . . . . . . . . . . . . . . . . . . . . 152 VI . Braginskii's Theory of Nearly Axisymmetric Fields . . . . . . . . . . 154 A . Lagrangian Transformation of the Induction Equation . . . . . . . 154 B. Nearly Axisymmetric Systems . . . . . . . . . . . . . . . . . . 155 C. Nearly Rectilinear Flows; Effective Fields . . . . . . . . . . . . . 157 D . Dynamo Equations for Nearly Rectilinear Flows . . . . . . . . . 158 E. Comments on the General Approach of Soward . . . . . . . . . . 160 F . Comparison between the Two-Scale and Nearly Axisymmetric 161 Approaches . . . . . . . . . . . . . . . . . . . . . . . . . VII . Analytical and Numerical Solutions ofthe Dynamo Equations . . . . . 163 A . The a* Dynamo with a Constant . . . . . . . . . . . . . . . . 163
119
120
H . K . Moffatt
164 165 166 168 168 170 171 174 176
B. a Dynamos with Antisymmetric a . . . . . . . . . . . . . . . . C. Local Behavior of a o Dynamos . . . . . . . . . . . . . . . . . D. Global Behavior of am Dynamos . . . . . . . . . . . . . . . . VIII. Dynamic Effects and Self-Equilibration . . . . . . . . . . . . . . . A. Waves Influenced by Coriolis Forces, and Associated Dynamo Action B. Magnetostrophic Flow and the Taylor Constraint . . . . . . . . . C. Excitation of Magnetostrophic (MAC) Waves by Unstable Stratification D. Mean Flow Equilibration . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . .
*
/J
I. Introduction
The existence of the magnetic field of the Earth, and its variation with time, presents a profound challenge to geophysics. This field, though influenced slightly by electric currents in the ionosphere, is predominantly of internal origin and is associated with a large-scale azimuthal current distribution in the liquid core of the Earth. It is well known (see, e.g., Hide and Roberts, 1961) that the temperature of the core is far above the critical value (the Curie point) at which permanent magnetization can persist. Moreover, in the absence of any regenerative action, the electric currents in the core would decay through ordinary resistive (ohmic) dissipation in a time of order 104-105years. Geomagnetic studies indicate, however, that the Earths field has existed in one form or another for at least 108years and is probably as old as the Earth itself, and further that, although the main dipole field exhibits random rapid reversals, a phenomenon reviewed by Bullard (1968), it remains at least quasi-steady between reversals for periods up to order 106 years, i.e., one or two orders of magnitude greater than the natural decay time. It is now generally agreed that this persistence of the Earths field can only be explained in terms of electromagnetic induction, whereby the electric currents that provide the field are generated by motion of the fluid in the core across the self-same field, which permeates the core region as well as the nonconducting exterior. The characteristic feature of such dynamo action is that the field is maintained exclusively by the action of the fluid velocity and without the help of any external source of field. This type of behavior is most simply illustrated in terms of the simple disk dynamo illustrated in Fig. 1. The electrically conducting disk rotates about its axis with angular velocity SZ, and a conducting wire makes sliding contact with the rim of the disk and with its axis, as shown; the wire is twisted into a circle in its passage from the rim to the
*
.mz
; -
*.
__
U
121
FIG. 1. The self-exciting disk dynamo; note particularly the concentrated shear at the sliding contact and the lack of reflectional symmetry of the system.
axis in such a way that any current flowing in the wire gives rise to a magnetic field with nonzero flux @ across the disk. The rotation of the disk in the presence of this flux generates a radial electromotive force, and since a closed circuit is available, a net current l ( t ) flows along the wire. The flux 0 then equals MI, where M is the mutual inductance between the wire and the rim of the disk, and I is given simply by
L dlfdt
+ RI = MRI,
(1.1)
c1
where L and R are, respectively, the self-inductance and resistance of the complete current circuit. Clearly, if R < MO, the current grows, and if R is maintained at a constant value, this growth is exponential. The state with I = 0 is then unstable to the growth of small electromagnetic disturbances. Such growth cannot, of course, continue indefinitely. The Lorentz force j A B (where j is the current distribution in the disk and B the magnetic field) provides a net resisting torque MI, and (1.1) must be coupled with the equation of angular motion of the disk (1.2) 0where C is the moment of inertia of the disk about its axis, and G the applied torque. If G (rather than R) is kept constant, then as I increases, R decreases until an equilibrium is reached in which
G = M12,
C dR/dt = G - M12,
= RIM.
(1.3)
_I
Note that the angular velocity of the disk in this equilibrium state does not depend on the applied torque! This simple dynamo relies for its success on the carefully contrived path that the current is forced to follow. The conductor (disk + wire) occupies a region of space that is not simply connected, a feature that is of course not shared by the conducting core of the Earth. There are, however, two features of the disk dynamo configuration that deserve particular emphasis, because
122
H. K . Moflatt
these features do recur in the fluid context and are both intimately associated with successful dynamo action. First, there is a region of concentrated shear at the sliding contact on the rim of the disk; the counterpart of this in the fluid context is differential rotation, which plays an important part in generating toroidal magnetic field from poloidal magnetic field (see Section 1 1 D Second, the configuration lacks reflectional symmetry in that the sense 1,) . of twist of the wire in Fig. 1 bears a very definite relation to the sense of the angular velocity of the disk: if the twist is reversed, then MQZ in (1.1) is replaced by - MRZ, and rather than dynamo action we have accelerated decay of any transient current in the wire. This lack of reflectional symmetry also has its counterpart in homogeneous fluid systems. The simplest measure of the lack of reflectional symmetry of a localized fluid motion u(x) is its helicity
,
,
-,
H =
.c
( V ~ Ud3x, )
(1.4)
a quantity that admits interpretation in terms of the degree of linkage (or " knottedness ") of its constituent vortex lines (Moffatt, 1969). We shall describe in the following sections (particularly Section V) how the presence of helicity is of vital importance for the process of regeneration of poloidal field from toroidal field, i.e., for the closing of the dynamo cycle that makes field regeneration a reality. The Earth is, of course, not the only celestial body that exhibits a significant large-scale magnetic field. Among the planets, Jupiter, Mars, and Mercury are now known to have this property also; the radii, rotation rates, and dipole moments of these planets in comparison with the Earth are displayed in Table 1 (from Dolginov, 1975). A theory that successfully explains the Earth's field may clearly have relevance in the context of these
TABLE 1
COMPARATNE FIGURES THE EARTH,JUPITER,' MARS: AND MERCURY~ FOR Radius R (km) Earth Jupiter Mars Mercury
a
Dipole moment p (G km3)

8.05 x 1.31 x 2.47 x 4.8 x 10" 1015 107 107
M~
(G)
3.11 x 10-I 3.61 6.36 x 10-4 3.31 x 10-3
Rotation period (days)

1 0.415 1 58
6371 71,351 3386 2439
* Dolginov et al. (1973).

Ness et al. (1974).
Warwick (1963); Smith et al. (1974).
123
other planetary fields. The internal constitution of Mercury, Mars, and Jupiter is, of course, largely a matter of speculation at present; it may be that detailed observation of the surface magnetic fields of these planets will in the long run provide (via theoretical argument) information about their interiors. * In the case of Jupiter, it has been argued (see, e.g., Hide, 1974) that the core consists of a liquid alloy of metallic hydrogen and helium under high pressure and that this core provides the seat of magnetohydrodynamic dynamo action. The magnetic field of the Sun (which is believed to be typical of cool stars with convective outer envelopes) is much more complex in its structure a n d behavior than that of the Earth, and it too is widely (though not universally) believed to be the result of dynamo action involving the two features, differential rotation and motions lacking reflectional symmetry, described above. Paradoxically, although the Sun is certainly remote as compared with the Earths liquid core, we have much more detailed information about its magnetic field, simply because it may be detected and measured at its visible surface, i.e., at the surface of the convective region where, from a magnetohydrodynamic point of view, all the interesting action takes place. In the case of the Earth, we have no prospect or hope of any direct measurement of the magnetic field, or indeed of any other quantity, either in the liquid core or on its surface, and we must make do with what we know of the field on the surface of the solid Earth, a pale shadow of the field of the deep interior, and a dim and distant reflection of the fluid turmoil in .f that deep interior which is at the heart of our problem. The Suns field is characterized by active regions and by a weak general field near the north and south poles of its axis of rotation. Active regions are associated with strong upwelling from the convective envelope with an associated vertical stretching of any magnetic field lines that are convected upward. When this stretching is particularly localized and intense, the strong e e r t i c a l field that is created (of the order of thousands of gauss) can locally suppress thermal convection; the resulting decrease in heat transport leads to a local cooling of the surface; radiation from this local region (of the order of hundreds of kilometers in horizontal extent) is largely suppressed, and in consequence it is seen from the Earth as a dark spot on the surface of the Sun. Such sunspots occur in pairs, roughly along a line of latitude, but with the leading spot (i.e., that to the East) slightly nearer the equatorial plane. Sunspot activity has been followed for more than 300 years and is known to - b . follow a roughly periodic cycle with half-period of about 11 years. At the beginning of a sunspot cycle, pairs of spots appear within the band of latitudes about 30from the equatorial plane, with statistical symmetry about -0 this plane, first at the higher latitudes only, then gradually over a wider band of latitudes that drifts, as the cycle proceeds, toward the equatorial plane. In
124
H.K. MofSatt
any pair of sunspots, the vertical magnetic field is positive in one and negative in the other; if positive in the westerly spot, the pair has positive polarity, otherwise negative. In any half-cycle of 11 years, all sunspot pairs in the northern hemisphere have the same polarity, and all in the southern hemisphere have the opposite polarity. In the following half-cycle, these polarities are reversed. The weak polar field of the Sun also follows a somewhat irregular periodic evolution with approximately the same period as that of the sunspot cycle. The field was first measured by direct magnetograph measurements in 1952 (Babcock and Babcock, 1955) and it has been followed closely since that date. The field around the north pole reversed in 1958 and again in 1971;the field around the south pole reversed in 1957 and again in 1972! At each reversal, for about one year, the fields at north and south poles therefore had quadrupole rather than dipole symmetry about the equatorial plane. The reversals apparently occur during that part of the sunspot cycle when sunspot activity is at its maximum. These observations are compatible with the following qualitative picture (essentially conceived by Parker, 1955a):the global magnetic field of the Sun includes poloidal and toroidal ingredients that are not steady but vary periodically in time, with period approximately 22 years. The poloidal field can be observed and has the polar reversal behavior described above; the toroidal field is contained in some way beneath the visible surface of the sun and cannot be directly detected. This toroidal field is coupled with the poloidal field and in a typical half-period drifts like a wave from polar regions toward equatorial regions, intensifying as it progresses. When this field reaches a certain critical level of intensity, local upwelling instabilities may develop in which ropes of toroidal flux are stretched vertically upward, breaking through the visible surface of the sun and forming sunspots as described above. As the buoyant fluid rises through several scale-heights, it expands due to the decreasing ambient pressure; as a result of the tendency to cn. oserve angular momentum, the rising blob develops a rotation (the sense of rotation being such that it has negative helicity in the northern hemisphere, positive in the southern): hence the twist of the sunspot pair from the original line of latitude of the underlying toroidal field. As the periodic evolution proceeds, the toroidal fields of opposite signs from the two hemispheres interpenetrate and annihilate each other in the equatorial zone and the sunspot activity in consequence dies away. The process then repeats itself, the toroidal field again growing from polar to equatorial regions (but with a complete change of polarity from one half-cycle to the next). What part does the weak poloidal field play in this process? It just must be present, for otherwise the dynamo cycle cannot proceed. On the one hand, all the little eruptions over the solar surface generate a field with a
. 1
.
-,
a--
Irr
125
radial component (i.e., poloidal field); since the surface eruptions are most intense in equatorial latitudes, one might expect the poloidal field to be most evident in these regions also. However, poloidal field can be redistributed by large-scale meridional circulation in the convective zone and this must presumably play a part in sweeping poloidal field back to the polar regions. Meridional circulation also tends to generate differentialrotation (conservation of angular momentum again) and this differential rotation is the means by which the toroidal field is regenerated from the poloidal. These complicated interactions may seem far removed from the simplicity of the disk dynamo described at the outset; yet the two features-differential .rotation and lack of reflexional symmetry-appear in the solar context as vital ingredients in its periodic behavior ; the lack of reflexional symmetry appears in the rising, twisting blobs, described by Parker (1955a) as cyclonic events, and directly responsible for sunspot formation. The above description is, of course, purely qualitative and suggestive. In the sections that follow, we shall endeavor to show how the various physical ideas implicit in the description may be placed on a secure mathematical foundation, and to relate the various approaches to dynamo theory that have made progress in this direction over the last 20 years. The reader who wishes further background material in the terrestrial and solar contexts may consult a number of review articles that have appeared in recent years (Parker, 1970a; Roberts, 1971; Weiss, 1971, 1974; Roberts and Soward, 1972; Vainshtein and Zeldovich, 1972; Moffatt, 1973; Gubbins, 1974) and the very extensive detailed references that these articles contain.
11. Magnetokinematic Preliminaries
A. IDEALIZATION KINEMATIC OF THE DYNAMO PROBLEM

Suppose that fluid of uniform electrical conductivity 0 is confined to a simply connected region of space V inside a closed surface S, and suppose that the region P exterior to S (extending to infinity) is nonconducting. Let u(x, t ) be the velocity field in V, satisfying n-u=O on S, (2.1)
and let p(x, t ) be the density field satisfying the equation of mass conservation
+/at
+ v - (pu) = 0.
(2.2)
126
H . K . MofSatt
*
For many purposes it will be sufficient to restrict attention to incompressible fluids of uniform density for which
V*B=O,
poj = V A B =poo(E + UAB),
(2.4)
(2.6) aB/at = - V A E , where po = 4n x 10- S.I. units. In p, where j = 0, B is determined by
V*B=O, VAB=O. (2.71 Moreover, both normal and tangential components of B must be continuous across S, i.e., [n.B]? =0, [nhB]? = O on S, (2.8) where n is the unit outward normal on S. Finally, we require that B be without singularities in I/ and in p, and that there should be no sources of magnetic field at infinity; this means that B must be at most dipole, i.e., 0 ( ~ - 3 ) as = l X j + CO. Elimination of j and E from (2.4)-(2.6) gives the well-known induction equation (which holds in V), (2.9) where 1= (po 0)- is the magnetic difusiuity of the fluid. For given U, we n. wish to explore the evolution of the field B as determined by (2.7)-(2.9) ad the subsidiary conditions mentioned. A simple measure of the field level is the total magnetic energy M(t) = (2p0)- J
V+P
aB/at = v A (U A B) + PB,
B2 d 3 x .
(2.10)
If, for given U, M ( t )+ 0 as t + CO, then the motion U does not act as a dynamo. If M(t) f ,0 as t -,CO, then the motion does act as a dynamo, there being ultimately sufficient rate of generation of magnetic energy by fluid motion to counteract the natural decay of magnetic energy due to ohmic dissipation associated with the finite conductivity of the fluid. In a complete theory that takes account of the dynamics of the fluid motion, U is, of course, constrained to satisfy the Navier-Stokes equation
.c
v.u=o. (2.3) Let j(x, t), B(x, t), and E(x, t) denote electric current, magnetic field, and electric field, respectively. Neglecting displacement current (certainly valid for phenomena on the long time scales considered), the equations relating j, B, and E in I/ are
p=po,
,*
*
127
(with Coriolis forces, Lorentz forces, buoyancy forces, etc., included if the context so requires). In a purely kinematic approach at the outset, it proves useful to widen the scope of the investigation and to imagine that U is any kinematically possible velocity field (without dynamical restriction); the influence of dynamic constraints, which will be considered in Section VIII, is more easily comprehended after close investigation Qf the kinematic problem.
B. MAGNETIC FIELD REPRESENTATIONS

1. Poloidal and Toroidal Decomposition Any solenoidal field B may be expressed as the sum of a poloidal ingredient B, and a toroidal ingredient B,, where B, = V A V A (xS(x)), B, = V A ( x T ( x ) ) ;
(2.1 1) (2.12) (2.13)
S and T are the dejning scalars for these fields. Note that
VAB, = V A V A (X T )
is a poloidal field with defining scalar T, while VAB, = -v2(vAXS) = -VA(XV2S)
is a toroidal field with defining scalar -V2S. Note further that x * l&T. = 0, i.e., the lines of force of the B, field lie on spheres r = const, as do the fines of force of V A Bp, a property that makes the representation particularly useful when problems with spherical boundaries are considered.
2. Axisymmetric Fields
A field B is axisymmetric about an axis Oz when its defining scalars Sand T are independent of the azimuth angle # about Oz. If S = S(r, O), T = T(r, 6), where 8 is measured from 02, then
where
x = -r
?
-w
sin 6 aSJa6,
B, = -8TJa6.
(2.15)
x, thejuxfunction, is the analog of the Stokes stream function for solenoidal velocity fields, and the lines of force of the B,field are given by x = const.
The B, field may also be expressed in the form B, = V A = x/r sin 8, and 4 is a unit vector in the I$ direction.
A
(A&), where
128
H. K. Mofatt 3. Two-Dimensional Fields
It is frequently illuminating to consider configurations in which B depends only on two cartesian coordinates, say x and y. In this case, the representation analogous to the above is B = B, B,, where now
and the BP lines are given by A = const. C. ALFVBN'STHEOREM WOLTJER'S AND INVARIANT It is an immediate consequent of (2.4)-(2.6) that, if @(t) is the flux of B across any surface spanning a closed curve C ( t ) that moves with the fluid, then
0
(2.17)
d@/dt =
-$
(E + U
C(t)
B) dx = fC(0
a-'j
dx,
so that, in the perfect conductivity limit (a = CO), @ is constant for any material curve C(t).It follows that in this limit B lines are frozen in the fluid, and in an incompressible flow, stretching of the B lines leads to proportionate intensification of the B field. Closely associated with the " frozen-field " concept is the invariance of the integra1
H - A*Bd3x,
4"
(2.18)
(Woltjer, 1958). Here A is the vector potential of B, i.e., B = V A A, and it is supposed that B is a localized field so that the integral exists. The volume I/ in (2.18) may be any volume bounded by a material surface S on which B * n = 0. The interpretation of H M is identical with that for the helicity integral (1.4) (which is likewise constant whenever circumstances are such that vortex lines move with the fluid), i.e., HM is a measure of the degree of topological complexity of the field B within the surface S-and this measure cannot, of course, change when the B lines are frozen in the fluid. We may note in passing that the solution of (2.9) in the limit 0 = CO (i.e., I = 0) may be expressed in Lagrangian variables in the form (due to Cauchy) (2.19) B ~ ( xt) = Bj(a, 0) a x i / a a j , , where x(a, t) is the position at time t of the fluid particle that was at position a at time t = 0.
0
I
129
D. NATURAL DECAY MODES AND FORCE-FREE FIELDS
If U is steady, i.e., U = u(x), then the problem (2.7)-(2.9) admits solutions proportional to exp( -pt), where possible values of p are determinate as eigenvalues of the problem. If, for all these values, Re p > 0, then B inevitably decays with time, while if for any eigenvalue Re p < 0, the corresponding field structure (the eigenfunction) grows exponentially in time until Lorentz forces modify the velocity field (cf. the rotating-disk situation discussed in the introduction). If p = pr ipi then the condition pr = 0 is critical in that it determines the onset of dynamo action for the corresponding field structure B,(x). If, when pr = 0, it also happens that pi = 0, then the resulting mode is steady under critical conditions; this is the sort of behavior that we look for in the context of the Earth's magnetic field, which is steady (with weak fluctuations) over very long periods. If, on the other hand, p i # 0 when p, = 0, then the resulting mode is oscillating under critical conditions; this type of behavior would be relevant in the solar context. Of course, when U 0, all the p's are real and positive; in the important case when the volume V is spherical with radius R, the eigenvalues are given by pnq = AR- 'x:~, (2.20)
where xnqis the qth zero of the Bessel function J,+ 1/2(x).The structure of the corresponding fields for r c R are given, in the notation of Bullard and Gellman (1954), by
SY + iSy = V A V A [ ~ ~ - ~ ' ~ J , + ~ , ~ ( A - ' ~ , ~ ~ ) ~ (2.21) ~~],

(2.22) The field Sy matches to a dipole field in the exterior region r > R, the field ST matches to an axisymmetric quadrupole field, and so on. From (2.20) the time scale of decay of these modes of simple structure is O(R'1- ') (as can, of course, be anticipated from dimensional analysis). The natural decay modes are closely related to field structures that are force-free, i.e., for which the Lorentz force j A B everywhere vanishes. Such fields arise naturally in the context of the kinematic dynamo problem, and it will be useful to set out some of their properties here. First, for such fields there exists a scalar field K(x) such that p;'j = V A B = K(x)B,
1
V
(2.23) (2.24)
and, since V j = V B = 0, it follows that B-VK=O and j-VK=O, i.e., B lines and j lines lie on a surface K = const.
130
H. K. Moflatt
The simplest example of a force-free field, with K constant everywhere, is (in cartesians) B = Bo(sin Kz, cos Kz, 0 ) ; (2.25)
the property V A B = KB is trivially verified. The B lines lie in planes z = const and rotate in a left-handed sense with increasing z. The vector potential of B is just A = K-B, so that A B = K-B = K-B$. (2.26)
The magnetic helicity density A B is thus uniform; the field structure (2.25) has maximal helicity (Kraichnan, 1973). This and other similar examples have the property that the j field extends to infinity. There are in fact no force-free fields for which j is confined to a finite volume and B is everywhere continuous and O(r- ) at infinity (see, e.g., Roberts, 1967, p. 109). It is, however, possible (Chandrasekhar, 1956) to construct force-free fields in a sphere V that match smoothly onto currentfree fields in the exterior region P that do not vanish at infinity: let
S(r, e) = Ar-2J3i2(Kr) COS 8,

and B = VA(XS) K-VAVA(XS) + for r
(2.27)
-= R.
(2.28)
Then it may be readily verified that, since S satisfies the Helmholtz equation (V + K)S = 0, the field (2.28) does satisfy V A B = KB. However, since j n = 0 on r = R, and since B is parallel to j, we must also satisfy B * n = 0 on r = R (a condition that the decay modes do not satisfy) and this requires that J3/2(KR) = 0. The exterior field in P is purely poloidal and is given by B = K- V A V A (xS), where
-.
S = - Bo(r - R3/r2) cos 8,
B, = 3AR-
I2
dJ,,,(KR)/dR, (2.29)
the latter condition ensuring smoothness across r = R.
111. Convection, Distortion, and Diffusion of B Lines

c
In this section, we consider certain particular solutions of the induction equation (2.9) when U(.) is prescribed and steady. The behavior is strongly dependent on the order of magnitude of the magnetic Reynolds number R, = U, l/A [where uo and I are, respectively, velocity and length scales characteristic of U(.)]. Moreover, great care must be exercised when the two
,--

*
131
limiting processes R, - CO and t - CO are considered-the solution may , , depend in a most sensitive manner on the order in which these limits are taken.
A. BALANCE STRETCHING DIFFUSION A OF AND IN MAGNETIC FLUXROPE

Equation (2.9) represents the evolution of B under the joint action of the stretching of magnetic field lines and of their diffusion relative to the fluid. A well-known steady solution of the equation in which these effects exactly balance exists when U is the uniform extensional straining motion given by
U = (-ax, -ay,
2az),
> 0.
(34
The steady solution of (2.9) is then
. )
B = (0, 0, Bo exp( -a/A)(x y)), (34 representing a flux rope of gaussian structure aligned with the z axis. The field (3.2), in fact, provides the asymptotic solution of (2.9) as t -,CO (with A fixed and nonzero). The flux in the rope is nBOA/a, that for given flux, B , so becomes very large when A is very small, under this type of persistent stretching. This type of motion may be expected (locally) to generate the strong vertical magnetic fields observed in sunspots, as described in the introduction.
B.
FLUX EXPULSION FLOWS BY WITH
CLOSED STREAMLINES
.
-*
The phenomenon of flux expulsion was first explicitly considered by Parker (1963) and Weiss (1966). Suppose we have a steady two-dimensional incompressible flow with stream function @(x, y) and suppose that the fluid is permeated at time t = 0 by a magnetic field that is uniform and in the plane of the flow. For t > 0 the flow distorts the field and diffusion, of course, also influences its behavior. If R, Q 1, diffusion dominates and the field perturbations remain small [in fact, O(R,) relative to the initial field]. If R, % 1, the picture is very much more complicated. During an initial phase whose duration is O(Rk/))I/u, (I and U, being the scales introduced above), diffusion is negligible and the field perturbations grow in intensity to O(RA/) times the initial field. There is then an intermediate stage, which has been studied in a particular case by Parker (1966), and whose duration is O ( R ~ / ) l / u , during which diffusion causes the breaking off of closed flux , loops in regions of closed streamlines; these flux loops slowly decay and disappear and there is an associated net reduction in the total flux threading
132
H . K . Moflatt
the region of closed streamlines. Finally, the field settles down to a steady state, in which the flux across any region of closed streamlines is exponentially small. The difference between limtdmlimA+oand limA+o limtdm is quite striking in this context: the former procedure gives a field whose energy density increases without limit (as t 2 ) ; the latter procedure gives a steady field whose energy density is generally less than that of the uniform field that we started with! The following simple proof that the flux across any region of closed streamlines must vanish (under the second limiting procedure) appears to be new. Using the representation (2.16a) for the magnetic field, (2.5) may be expressed in the simple form (3-3) It is supposed here that there is no applied electric field, so that E, = -aA/at. Under steady conditions then, V * (uA) = U V A = IV2A, (3.4) and in the limit I + O , U V A = 0, so that A is constant on streamlines, or equivalently A = A($). We now adapt the argument of Batchelor (1956) (as applied to the vorticity equation) to the present context: let C be any closed streamline, and integrate (3.4) over the area enclosed by C. Since n U = 0 on C (where n is normal to C), the left-hand side integrates to zero. This focuses attention on the effects of diffusion when I is small but not quite zero. The right-hand side, on integration, gives DA/Dt z aA/at + U * V A = IV'A.
0 = I n V A ds = I dA/d$
C
+
$ a $ p n ds = I K c dA/d$,
C
(3.5)
where K c is the circulation around C. It follows that in the steady state (which may, of course, take a long time to attain) dA/d$ = 0, i.e., A = const, i.e., B = 0 throughout the region of closed streamlines. A net flux of field across, say, a periodic array of eddies with closed streamlines cannot, of course, be simply eliminated by this mechanism. What happens, as is clearly demonstrated in the numerical solutions of Weiss (1966), is that the flux is concentrated into sheets of thickness O(R;''2) at the boundaries between adjacent eddies. A horizontal row of eddies will concentrate vertical flux in this way at the vertical cell boundaries, whereas any horizontal flux will be expelled to the regions above and below the eddies. The above proof can be simply adapted to cover the corresponding axisymmetric situation when steady meridional circulation acts on an axisymmetric poloidal field. Again, if the relevant magnetic Reynolds number is large, the field is ultimately excluded from any region of closed
--
Generation o Magnetic Fields by Fluid Motion f

*
133
streamlines. This result has important implications for dynamo theory: meridional circulation that is weak can be conducive to efficient dynamo action, but meridional circulation that is too strong merely expels poloidal field from regions of closed streamlines (i.e., from the whole fluid region for an enclosed flow) and this effect is bound to be counterproductive, as indeed demonstrated in the numerical studies of P. H. Roberts (1972).
c. TOPOLOGICAL PUMPING OF MAGNETIC FLUX

A rather fundamental variant of the flux expulsion mechanism has recently been discovered by Drobyshevski and Yuferev (1974). This study was motivated by the observation that in steady Benard convection between horizontal planes, fluid rises at the center of the convection cells and falls at the periphery; the regions of rising fluid are therefore separated from each other, whereas the regions of falling fluid are all connected. If the fluid is permeated by a horizontal magnetic field, then a field line near the upper plane will be distorted to lie entirely in a region of falling fluid and will therefore tend to be convected downward, a tendency that may be resisted to some extent by diffusion. A field line near the lower plate cannot be distorted to lie everywhere in the disconnected regions of rising fluid and cannot therefore be convected upward. One would therefore expect a net tendency for flux to be transported toward the lower plate; the Benard layer should act as a valve, allowing horizontal flux to pass downward but not upward. The particular velocity field chosen by Drobyshevski and Yuferev to demonstrate. this effect is (in dimensionless form)
U = (-sin
x ( l + f cos y ) cos z,
- (1
+ f cos x) sin y cos z, (cos x + cos y + cos x cos y ) sin z),
.
P
(3.6) for which the cell boundaries are square; the more realistic choice of hexagonal cell boundaries would complicate the analysis without greatly adding to the insight provided. The magnetic field in the absence of fluid motion (or equivalently at zero magnetic Reynolds number) is taken to be ( B o ,0, 0), i.e., uniform in the x direction, and it is supposed that the boundaries z = 0, II are perfectly conducting so that the flux nBo is trapped in the gap between them. The steady solution of (2.9)may be obtained as a power series in R, when R , 4 1; what is of interest is the average of this field over the horizontal plane, which turns out to have the expansion 7Ri B ( z ) = Bo 1 + COS 22 + 7 COS z - 3 COS 32) + O(Ri) (28 4811 240n (3.7)
134
H . K. Mofatt
The asymmetry about z = 7c/2 appears at the O ( R i ) level, and this term indeed shows the expected increase in flux in the lower half of the gap. This phenomenon is of potential interest in the solar context. Convection in the Suns outer layers is turbulent, but there are also fairly stable and persistent large-scale structures, reminiscent of Benard cells, that survive even in the presence of this turbulence. These cells do show a preferred tendency for fluid to rise in discrete regions and to fall in connected regions, and any toroidal flux permeating this region will in consequence tend to be transported downward. The relevant magnetic Reynolds number is R,, = U, 1, /A, where U, and I, are scales characteristic of the large-scale structures and 1 is an effective eddy diffusivity associated with the small, scale turbulence (see Section V); R,, will probably be of order unity (even though the magnetic Reynolds number based on molecular diffusivity is very large). The phenomenon of flux pumping has been further investigated by Proctor (1975), who has demonstrated that when R, < 1, pumping can occur even when the topological distinction between upward- and downwardmoving fluid is absent. Proctor analyzes the effect of two-dimensional motions in detail and shows that a lack of geometrical symmetry about the midplane is sufficient to lead to a net transport of flux either up or down; e.g., if (w3)# 0, where w is the vertical velocity at the midplane and the angular brackets denote a horizontal average, then there will be a net transport, which Proctor describes as geometrical (as opposed to topological) pumping. When R, % 1, however, he demonstrates that this type of twodimensional geometrical pumping is almost nonexistent, whereas in this limit the Drobyshevski and Yuferev mechanism may be expected to be most effective (although no detailed analysis has yet been carried out). D. GENERATIONTOROIDAL OF FIELD DIFFERENTIAL BY ROTATION We conclude this section with a brief discussion of the process by which toroidal field is generated from poloidal field by differential rotation. Physically it is clear that differential rotation about an axis will tend to distort the field lines of an initially poloidal axisymmetric field B, if the angular velocity o varies along the field lines. In fact, if U = w(s, z)k A x, where (s, 4, z) are cylindrical polar coordinates and k a unit vector along Oz, and if B = B,(s, z ) + BJs, z, t)&, then the 4 component of (2.9) is
aB,/at = S(B,
-.
- v)w +
~ ( V Z s - 2 ) ~ ~ .
(3.8)
For the moment we shall suppose that B, is maintained steadily by some unspecified mechanism. As expected, it is the gradient of o along B, lines that gives rise to generation of toroidal field. If 1is small, and if initially
-_
135
B4 = 0, then B4 grows linearly with time until I B, I = O(R,) I B, 1, at which stage a steady state is established. There is no question of flux expulsion in this case since B, is axisymmetric (if B, included any nonaxisymmetric ingredients then these would be expelled). The ultimate steady solution of (3.8) may be easily obtained if Bp and o are prescribed. For example, if B, = B, k where B , is uniform, and if o = o ( r ) where r2 = s2 + z2,then the steady solution of (3.8) is
B, = -4B0(h3)-1 sin 6 cos 6
jorrfw(rl) dr,.
(3.9)
Note that B4 as given by this solution is antisymmetric about the equatorial plane 13 = a/2, and that if o ( r , ) decreases sufficiently rapidly as rl -+ CO for the integral in (3.9) to converge, then B, = O ( r - 3 ) as r + CO. . Contrast the situation when B, is an irrotational field with uniform gradient, i.e.,
B, = C( -2si,
+ zk),
(3.10)
where is is a unit vector in the s direction. The steady solution for B4 then has two ingredients proportional to sin 6 and dP,(cos 0)/86, but the former ingredient dominates for large r, and again provided w(rl) decreases , sufficiently rapidly as r , - CO, the asymptotic behavior of B, for large r is
9
B4 (2C sin 6/31r2)
j r f w ( r , )dr,.
0
(3.11)
This field is symmetric about 6 = 4 2 , and O ( F 2 )at infinity. In general, therefore, if a poloidal field B, is weakly varying in a region of differential rotation, then it is the local gradient of the field (rather than the local field itself) that determines the toroidal field generated at remote points when conditions are steady.
IV. Some Basic Results
A. COWLINGS THEOREM RELATED AND RESULTS
The impossibility of steady maintenance of an axisymmetric magnetic field by motions axisymmetric about the same axis (Cowling, 1934) is so well known as hardly to require comment here. The modifications and extensions of the theorem are numerous, but all reflect the basic fact that axisymmetric meridional circulation can redistribute poloidal flux but cannot , systematically regenerate it. The equation for the flux function ~ ( r6) under
136
H . K. Mofatt
axisymmetric conditions [analogous to (3.35)] is
axlat + up vx = A D Z ~ ,
where up is the meridional velocity (assumed axisymmetric) and D2 the Stokes operator. The structure of this parabolic equation essentially ensures (Braginskii, 1964a) that Vx everywhere tends to zero. Even if 1 is nonuniform but satisfies merely the natural condition up V 1 = 0, standard manipulation of (4.1) and the relevant boundary conditions leads to this same conclusion (yet another minor extension of Cowlings celebrated result !). Of course, when JVxI and so lBpl have reached a negligibly weak level, the toroidal field B6 must likewise decay, again essentially because of the parabolic structure of (3.8). An interesting variation on Cowlings theorem has been claimed by Pichakhchi (1966). This is that steady dynamo action is impossible if the electric field E vanishes everywhere. (In the axisymmetric case with B, = 0, this is just a rewording of Cowlings theorem since E E 0 in this situation under steady conditions.) If a steady dynamo with E E 0 were possible, then the field B would [from (2.5)] satisfy B (V A B) = 0, i.e., it would have zero helicity everywhere and a correspondingly simple topological structure. The fact that this is apparently not possible is of course significant. There is also a counterpart of Cowlings theorem for two-dimensional velocity and magnetic fields (Zeldovich, 1957; Lortz, 1968). In this case, (3.3) is the governing equation, and by standard manipulations, provided A = o(r- ) at infinity,
(d/dt)
JJ A2 dx d y = -21
ss
(VA) dx dy.
(4.2)
It follows that a steady state is possible only if VA = 0, i.e., only if B, = By = 0. Again, this removes the source of any possible regeneration of B, , which must also therefore vanish in a steady state. If U and B are stationary random functions (with zero mean) of x and y , then (4.2) must be replaced by
d( A2)/dt = - 2( (VA)2), 1
(4.3)
where ( ) indicates averaging over the x-y plane. No matter how small 1 may be, this again implies ultimate decay of the magnetic energy density. In the limit 1 -+ 0, ( A 2 ) apparently remains constant; however, this holds only so long as ((VA)2) remains finite; in fact, ((VA)) increases (just as in the particular situation described in Section II1,B) until it is O(1- ); at this stage and the length scale of the field is 0(1/2) the inexorable decay of the field then sets in.
i
*-
Generation of Magnetic. Fields by Fluid Motion
137
The impossibility of dynamo action under purely toroidal motion (Bullard and Gellman, 1954; Backus, 1958) is also closely related to Cowlings theorem, in that it follows from the equation
(4.4) which may be derived from (2.9) together with V U = 0. If U is purely toroidal, then x * U = 0 and so x B inevitably decays everywhere to zero. The decay of Bp and B, is almost an immediate consequence. Busse (1975) has on the basis of Ea. (4.4) obtained a necessary condition (in terms of a lower bound on the poloidal velocity) that muscbe satisfied for successful dynamo action.
D(x * B)/Dt = (B V)(X * U)
+ 1V2(x - B),
B. ROTORDYNAMOS In view of the various antidynamo theorems described above, it was of course of crucial importance that the possibility of steady dynamo action in a fluid of uniform conductivity occupying a simply connected domain be unambiguously established for at least one kinematically possible velocity field (no matter how artificial from a dynamical point of view). Until this was done (Herzenberg, 1958) it was by no means clear that a master theorem proving the absolute impossibility of such dynamo action might not at some stage be proved. Herzenbergs dynamo consisted of two spherical rotors (i.e., quasi-eddies) embedded in a fluid sphere, the conductivity being uniform throughout. In fact, it is easier to comprehend the three-rotor problem (Fig. 2) considered? by Gibson (1968). Suppose that three spheres S,, S, , S , each of radius a, with centers located at the points (d, 0, 0), (0, d, 0), and (0, 0, d), where d 4 a, rotate with angular velocities (0, 0, -o), (-0, 0, 0), and (0, -0, 0), respectively, where o > 0; the conductivity D throughout the whole space is assumed uniform. Note immediately the lack of reflectional symmetry in this configuration. The principle of the dynamo is roughly as follows: suppose that there are nearly uniform fields of the form B, x (0, 0, B), B, z (B, 0, 0), and B, x (0, B, 0) in the neighborhoods of S , , S , , and S, , respectively. Then we have the possibility of a cyclic generation in which the toroidal field generated by rotation of S, (a = 1, 2) acts as the applied field B,,, in the neighborhood of S , + , , and the toroidal field generated by rotation of S , acts as the applied field B, in the neighborhood of S1.The subtlety of the problem derives from the fact that, as mentioned at the end of Section III,D, the fields generated by differential rotation in each case are determined to an important degree by the local field gradient as well as by the local field itself, and this has to be taken into account in the
.
.
-.
t The particular configuration considered here was also discussed by Venezian (1967).
138
H . K . MofSatt
FIG.2. The three-rotor dynamo of Gibson (1968). The spheres rotate as indicated and generate toroidal fields Be.which act as the applied poloidal fields for S (a = 1, 2, 3). ,
detailed calculation. The condition obtained by Gibson for steady dynamo action (adapted to this particular configuration) is
R , = oa2/3,= lO$
(d/~)~,
(4.5)
correct to leading order in the small parameter a/d. A working dynamo based on the interaction of two rotors has been constructed in the laboratory by Lowes and Wilkinson (1963, 1968). The rotors are cylinders inclined to each other and embedded in a block of material of the same conductivity, electrical communication between the rotors and the block being provided by a lubricating film of mercury. Not only was dynamo action demonstrated with this model [through observation of a sudden large increase in the local magnetic field when the angular velocities of the cylinders were increased beyond a certain critical value-analogous to that given by (4.5)], but also reversals of the field were observed when the dynamo was functioning in the fully nonlinear regime-an observation of the greatest interest in view of the known random reversals of the Earths magnetic field mentioned in the introduction. A further ingenious example of dynamo action associated with a pair of rotors has been analyzed by Gailitis (1970). In this case, the rotors are toroidal rather than spherical, and the velocity field is axisymmetric about the common axis of the two toruses. The magnetic field that is maintained is, however, nonaxisymmetric, and there is no conflict with Cowlings theorem.
a
( .
139
V. The Mean Electromotive Force Generated by a Random Velocity Field

A. THE TWO-SCALE APPROACH
Motion in the convective zone of the Sun is certainly turbulent, and any dynamo theory that fails to take account of this fact is hardly realistic. Likewise, motion in the core of the Earth almost certainly consists of a mean and a random ingredient. It is not clear whether the random ingredient is turbulence in the normal sense of the word, or rather a random field of waves influenced by Lorentz, Coriolis, and buoyancy forces; from a purely kinematic point of view, this distinction is not crucial, and we merely assume ( throughout this section that U& t ) is a stationary random function of both x and t with zero mean-effects of nonzero mean velocity will be considered in Section VI. We are primarily concerned with the evolution of the mean magnetic field, which may be supposed to have a characteristic length scale L large compared with the scale 1 that characterizes the background velocity field U; in the case of turbulence, 1 will be the scale of the energy-containing eddies (Batchelor, 1953), while in the case of random waves, 1 will be, say, the wavelength of the most energetic modes represented in the spectrum of U. Either way, we can average the induction equation (2.9) over a spatial scale intermediate between 1 and L to obtain (54 where B , = ( B ) , B = B, + b, and d = (U A b). This two-scale approach was introduced by Steenbeck et al. (1966) and has since had a revolutionary impact on the subject?. A series of papers by these authors, developing their approach to mean field electrodynamics has been collected together in English translation by Roberts and Stix (1971). The main problem, of course, analogous to the closure problem of turbulence dynamics, is to find a means of expressing 6 in terms of B, so that (5.1) may be integrated. The task is, however, easier because the basic equation for B is linear, albeit with a random coefficient. The equation for b, obtained by subtracting (5.1) from (2.9), is
aB,/at = V A d
+ 1V2B, ,
ab/at=V~(u~B,)+V~(uAb-(uAb))+1V~b, (5.2) and if we suppose that b = 0 at same initial instant t = 0, this establishes a linear relationship between b and B, , and so between 6 and B,; since the
t The concepts of a mean electromotive force and of an associated eddy conductivity were already present in earlier work (e.g., Kovasznay, 1960).
140
H . K . MojJatt
scale L of B , is very large, such a relationship may presumably be developed as a series
aBoj/aXk + Y i j k l a2Boj/aXk 8x1 + * , (5.3) the coefficients a i j , &, . . .,being pseudotensors determined in principle by the statistical properties of the U field and the parameter 1 (which, of course, plays an important part in the solution of (5.2) (pseudo because d is a polar vector, whereas Bo is an axial vector). It is important to note that these pseudotensors do not depend on B,; hence aij may be evaluated on the simplifying assumption that B, is uniform; then B i j k may be evaluated on the assumption that dBoj/axkis uniform, and so on. Attention in most investigations has been focused on the first two terms of (5.3) on the grounds that it is essentially a series in ascending powers of l/L, and subsequent terms are likely to have negligible effect when L is large. There are, however, considerable subtleties here, particularly when 1 is very small, and the possible influence of subsequent terms perhaps deserves investigation. For the present, however, we truncate the series (5.3) after the second term and investigate some of the consequences. First, suppose that the U field exhibits no preferred direction in its statistical properties; since a i j , / ? i j k , are essentially statistical properties of the turbulence, they must then be invariant under rotations of the frame of reference and must therefore take the form
bi = aij&j
Pijk
~
(5.4)
where a is a pseudoscalar and /? a pure scalar. Here the crucial role played by lack of reflectional symmetry comes into evidence. If the U field is reflectionally (as well as rotationally) symmetric in its statistical properties, then a (being a pseudoscalar that is not invariant under change from a rightto a left-handed frame of reference) must vanish. No such conclusion applies to the fl term. If the turbulence lacks reflectional symmetry, then the a term will in general be nonzero, and the relation between d and Bo becomes
0
-.
d=aB,-/?VhB,. In view of the assumed statistical homogeneity of the constants, and (5.1) becomes aB,/dt = ~ V A B + (1 P)VZB0 ,
(5.5) field, a and /? are

(5.6)
Hence fi plays the role of a turbulent diffusivity; it is to be expected that / > 0, although no general proof of this appears yet to be available. The a I term, on the other hand, has quite a novel structure (from the point of view of conventional electrodynamics) and is in fact of crucial importance for dynamo theory.
141
The average of Ohms law (2.5), incorporating ( 5 4 , becomes J, = a(E, aB, - BV A B,),
(5.7)
where Jo = (j), E, = (E), or equivalently,
ne = a 1 Bapo)( Jo = a,(E, aB,), (5.8) The a term therefore tends to drive mean current along the lines of mean magnetic field. In a spherical geometry, this effect in the presence of a toroidal field will generate a toroidal current, which acts as the source of a poloidal field. This therefore is the key to the means by which poloidal field may be regenerated from toroidal field by nonaxisymmetric random motions. The explosive character of Eq. (5.6) may be recognized very quickly if we suppose for the moment that our fluid fills all space and if we consider a magnetic field having an initial structure satisfying V A B, = KB, (i.e., one of the force-free fields of Section 11,D). For such a field, VB, = -K2B,, and so according to (5.6) the field will retain its spatial structure and develop exponentially like exp o t , where
o = aK
- ( A + B)K.
(5.9)
(5.10)
Clearly we have exponential growth (i.e., dynamo action) if

a~
a
=. (.a + p ) ~
(and clearly we must choose K to have the same sign as a in order to ensure this possibility). Provided K I is sufficientlysmall (i.e., provided the scale of B, is sufficiently large), condition (5.10) is satisfied. Dynamo growth is then assured for force-free modes of sufficiently large length scale. It is, of course, important to obtain an explicit representation of aij and Bijk in terms of the statistical properties of the U field; this stage of the problem is analogous to the statistical mechanics problem of obtaining expressions for the various transport coefficients in terms of the statistical substructure of the medium considered. In the present context the substructure is provided by the background random velocity field. Unfortunately, determination of tlij and Bijk is possible only in certain limiting situations; however, these limiting situations are in themselves of particular interest and will be considered in the following sections.
B. THESTRONG DIFFUSION LIMIT

0
If R , = U, l / A 4 1, where U , = (u2)/, then the diffusion term AV2b in (5.2) clearly dominates over the interaction term
G = VA(uAb - (UAb)),
(5.11)
142
H. K. Mofatt
which may be neglected to lowest order. The resulting equation
ab/at = V A (U AB,) + lV2b
(5.12)
is soluble by straightforward means, as recognized by Liepmann (1952) in a discussion of the spectrum of field fluctuations generated by turbulence in the presence of a uniform magnetic field. It is important now to distinguish between conventional turbulence whose characteristic time scale (the turnover time of the energy-containing eddies) is of order I/uo, and a random wave field whose time scale (the inverse of the dominant frequency) is determined by the relevant dispersion relation and is quite independent of l/u, . In the former case, 1 ab/at I is of the same order of magnitude as I G I and should, for consistency, be dropped also. Let us first determine aij in this situation. As remarked earlier, to do this we may suppose Bo uniform, and (5.12) becomes simply
1V2b = - (B,
V)U.
(5.13)
Let p(k, t) and q(k, t) be the space Fourier transforms of U and b, respectively. [The fact that p and q must be generalized functions (Lighthill, 1959) does not invalidate any of the operations that follow.] Then from (5.13),
- l k 2 q = -i(Bo
k)p.
(5.14)
The spectrum tensor Qij(k) is related to p(k, t) by

<pi@, t)p,*(k, t)> = @ij(k) a(k - k) ,
(5.15)
where an asterisk here and subsequently denotes a complex conjugate, and similarly for the spectrum tensor Tij(k) of the b field. The relation between these two spectrum tensors may be derived immediately from (5.14) (Golitsyn, 1960) in the form
Tij(k) = (lk2)-(B0 * k)2@ij(k).
(5.16)
More interestingly, the vector 8 may be obtained from (5.14) in the form
Bi= a i j B o j ,where
uij = i.cikll-
j k-kj@,,(k)
d3k.
(5.17)
The Hermitian symmetry of Okl(k) (Batchelor, 1953) guarantees that this expression for aij is real; moreover, the incompressibility condition in the form kiQij(k) = 0 may be used to show that aij [as given by (5.17)] is also symmetric (Moffatt, 1970a).
*-
143
For turbulence that is statistically invariant under rotations (i.e., showing no preferred direction), Qkl(k) takes the form
(5.18)
where E ( k ) ( 20 for all k) is the energy spectrum function, and F ( k ) the helicity spectrum function, satisfying
0 The Schwarz inequality

0
+U) (
.F
E ( k ) dk,
(U
V A U ) = Jo F ( k ) dk.
(5.19)
(PkAP*)I(IPI)(lkAPl)
(5.20) (5.21)
may be translated into spectral terms to show that, for all k,
IF ( k )I I 2kE(k).
It is clear that only the antisymmetric part of @kl contributes to (5.17), and substitution of (5.18) in fact gives aij = ad,, where
a = -(1/3A)
k-F(k) dk.
(5.22)
Thus a is expressed as a weighted integral of the helicity spectrum function. If F ( k )is either positive for all k or negative for all k, then a and (U V A U) have opposite signs; in this case an order of magnitude estimate of a is given from (5.19b) and (5.22) in the form
ax
-41- P ( U
o),
(5.23)
where o = V A U . The pseudotensor B i j k can be obtained by the same method and is likewise proportional to A- It follows on dimensional grounds that all components of B i j k have order of magnitude at most A- U: I = R i A; since R, 4 1, effects associated with Bijkare in this limit swamped by the molecular diffusion term AVB, in (5.1) and may reasonably be ignored. The conclusion that aij is symmetric does not persist if the calculation leading to (5.17) is extended to higher powers in R,. The result (5.17) may be regarded as the leading term of an expansion of the form
(5.24)
where the a$) are dimensionless pseudotensors that may in principle be determined by iteration based on (5.2); a$) can by this means be expressed as
144
H.K. Mofatt
an n-fold weighted integral of an ( n 1)th-order spectrum tensor. Krause (1968) has considered the question of convergence of this type of expansion and concludes that it does converge for all finite R,, although the indications are that the convergence may be extremely slow for large R,. There are further indications from parallel work by G. 0. Roberts (1970, 1972) on spatially periodic dynamos that a:;) is symmetric or antisymmetric according as n is odd or even; this conjecture requires further investigation in the turbulence context. may be decomposed into its symmetIngeneral, however, it is clear that ai, ric and antisymmetric parts:
aij = a$) - eijk5 .
(5.25)
Theeffect ofthe antisymmetricpart in (5.1) is to provide a term V A (V A B,) on the right-hand side, where V is a velocity (uniform in so far as the turbulence is homogeneous) that merely convects the large-scale field pattern. In conjunction with the a effect, the force-free modes considered in Section V,A would then propagate relative to the fluid with phase velocity V. If there is also a mean fluid velocity U, then the efective mean velocity as far as the magnetic field is concerned is U + V; this concept of an effective velocity is a crucial ingredient of Braginskiis (1964a) theory of nearly axisymmetric configurations (see Section VI). The appearance of an electromotive force with a component parallel to the ambient magnetic field [Eq. (5.5)] bears a simple physical interpretation, which was in fact given by Parker (1955a), who on the basis of inspired physical reasoning and heuristic argument anticipated the main lines of development of the subject by more than ten years! Consider the effect of a localized motion with positive helicity (a cyclonic event in Parkers terminology). Such a motion tends to generate an R-shaped loop in a line of force of an ambient magnetic field B, (Fig. 3a), and the loop is twisted so that its normal has a nonzero component in the original field direction.
(a)
(b)
FIG.3. Possible distortion of a line of force by a cyclonic event.

8
145
When the twist is of limited amount (as it certainly will be in the strongly diffusive situation considered above) the loop may be conceived as being due to a current antiparallel to B,; a random superposition of such motions may then be expected to generate a mean current Jo antiparallel to B, ,as implied by the result 8 = aB, with a < 0. This argument seems reliable both in the strong diffusion limit considered here and in the alternative situation considered by Parker when diffusion is weak but the events are so short-lived that the limited twist picture of Fig. 3a is applicable. In this weak diffusion limit, however, if the events are not short-lived, then a twist through 3n/2 (Fig. 3b) will give just the opposite effect from a twist through 4 2 , and this introduces some uncertainty into the sign of the effective value for a. Since the sign of a is of crucial importance for some of the dynamos described in Sections VI1,C and D it becomes important to have a reliable expression for a in the weak diffusion limit also (see Section V,D).
C. EVALUATION aij AND OF
Bijk FOR A RANDOM WAVE FIELD
For a random wave field, it is natural to use a double Fourier transform in both space and time variables for u(x, t):
*
u(x, t ) =
ss
ii(k, w)ei(.x - O t ) d3k d o ,
(5.26)
and similarly for b(x, t). The field has a random wave character (rather than a turbulent?character) if the velocity amplitudes in each constituent wave are small compared with their phase velocities, i.e., provided I k3wii(k, w )1 4 w/k. In this situation the interaction term G is again negligible in (5.2) and so b is determined by (5.12). Now, however, there is no reason to drop ab/at; moreover, (5.12) should be valid for all values of I, both small and large, the case when 1 is small being now of greater potential interest. Note that in the limit of infinitesimal wave amplitudes, the constituent waves in (5.26) are noninteracting, with a dispersion relationship (that will be determined by dynamic considerations) w = w(k). For finite amplitude waves, however, nonlinear interactions will provide some forcing of waves at nonresonant frequencies, and w need not then be restricted by a dispersion relation; there may of course also be some extraneous forcing of waves. Taking the Fourier transform of (5.12), with B,(x) a field whose second and higher space derivatives vanish, and constructing (U A b) as before leads
146
H.K . Mofatt
to the following expressions for aij and &k:

a.. = i h i k t IJ
(w2
+ 12k4)-'k2kj@kt(k, dk dw, 0)
12k4)-1(k0 -I- Ak2)
0)
(5.27)
pijk
= Re
Eiml
lj
(' 0
a (&
kk@tm(k,
- @jm(k,U)&
dk dw,
(5.28)
where now alm(k, is the double Fourier transform of the space-time w) velocity correlation tensor. Note that results for the strong diffusion limit w (Section V,B) may be recovered by replacement of aij(k, ) by @,(k) 6(0); in this limit, magnetic adjustment to velocity fluctuations becomes instantaneous. In the case when the wave amplitudes are isotropically distributed, (5.24) becomes aij = a a, where ,
m
a =+a.. =
-+I
dk(w2 + 12k4)-'k2F(k, w),
(5.29)
where F(k, w ) bears the same relation to Okt(k, ) as F ( k ) bears to a k l ( k ) w in (5.18). Similarly, (5.28) becomes P i j k = P E i j k , where
m m
p ='p. k E y. .k = gn 6 y
dk(w2+ L2k4)- 'k2E(k, w )
(5.30)
(Krause and Roberts, 1973; Roberts and Soward, 1975). Evidently / is posiI tive and again a has the opposite sign from F(k, w) if this function is single-signed. It is noteworthy that both expressions (5.29) and (5.30) vanish in the perfect conductivity limit I + 0 [unless there is a divergence of the integrals in the neighborhood of w = 0, a possibility that can certainly be discounted if Qij(k, w ) = O(02) as w + 0 .The reason is essentially that in this limit b 1 and U are in phase ineach constituent wave, so that ii A 6* makes no contribution to (U A b). Dillon (1975) has shown that, as far as a is concerned, this result holds even when effects of the interaction terms G are included: by expanding in powers of the amplitude of the constituent waves, Dillon shows by induction that, to all orders, aii = 0 when A = O! The vanishing of a and p in the limit I -,0 does, however, depend critically on the assumption of a well-established wave field, with no transient
147
effects present, and no zero frequency ingredients. If we adopt the alternative point of view and solve (5.12) with I = 0, B, uniform, and subject to the initial condition b(x, 0) = 0, we obtain (5.31) and so
(U A
b)i
= Eijp
jf<uj(xt ) du,(x,
0
z)/ax,)
U
dz B,,,
(5.32)
@and the relevant value of a when the field evidently

a= -
is isotropic and t large is
]om(u(x, t ) * V
u(x, t
- z)) dz = - -
3 jom
F(k, 0) dk, (5.33)
0
.
which need not vanish. So which expression is correct, (5.29) or (5.33)? If F(k, 0) # 0, then (5.33) is appropriate [and may in fact be obtained directly from (5.29) by replacing F(k, w ) by F(k, 0) and integrating with respect to 01; if F(k, 0) = 0 however (i.e., if the wave spectral density vanishes as w+O), then the expression (5.29) is appropriate with the implication [consistent with (5.33)] thata+OasI-+O. This distinction is clearly important since a factor I (or equivalently R; I ) in order of magnitude estimates of a makes a big difference when (as in the turbulent convection zone of the sun) R, % 1. The approach of Parker (1955a, 1970b) (see Section V,D) gives a result akin to (5.33) and corresponds to the limiting procedure liml+mlimA+o; approach of Braginskii the (1964a,b) gives a result more akin to (5.29) and corresponds to the inverse procedure 1imL+, liml+ .
D. EFFECT TUR~ULENCE WEAK DIFFUSION OF IN THE LIMIT, + 0 I For turbulence with typical time scale O(l/uo),there is no justification for the neglect of the interaction term G when I is small, and we must return to the exact Eq. (5.2). If we fix t 0 and let I -+ 0, then the Cauchy solution (2.19) is valid and we may construct
=-
bi = (U
b)i
= ( u A B ) ~ Eijk(vj(a,t)B,(a, 0) =
dxk/8a,),
(5.34)
where v(a, t ) = u(x(a, t), t). If further we adopt the initial condition b(x, 0) = 0, then B,(a, 0) = B,,(a, 0); and with the usual artifice that aijmay
148
H . K . Moflutt
be calculated by assuming Bo uniform, we obtain immediately
This tensor has a superficialresemblance with the diffusion tensor D i j ( t )for a scalar field (Taylor, 1922):
D i j ( t )= [<.,(a, t)uj(a, z)) dr.
0
(5.36)
There are, however, several difficulties in the interpretation of ( 5 . 3 9 which may be catalogued as follows: (i) If the turbulence is statistically stationary in space and time, then the integrand in (5.36) is a function of t - z only and the integral certainly converges as t -+ CO. The integrand of (5.35) may, however, depend on t and t - T independently (since auk/8u, is not a stationary random function of z-see Lumley, 1962) and there is no obvious guarantee that the integral , converges as t - 00. (ii) If the integral (5.35) diverges or oscillates as t - CO (as is not implausi, ble bearing in mind the discussion about loop twisting at the end of Section V,B), then the effects of weak molecular diffusion must in some way be reincorporated to force convergence in a time of order 12/I.This is extremely difficult, although possibly the approach of Saffman (1960) for the corresponding scalar problem might succeed. (iii) Even if the integral (5.35) does converge, there is no absolute guarantee that it gives a good approximation to the relevant value of a when Iz is small but nonzero. We have seen in Section V,C that expression (5.33) (which is certainly a convergent integral) may be quite misleading when R , is large but finite and t - CO. The same may be true in the present , context in which Lagrangian (rather than Eulerian) correlation tensors make a natural appearance. In the linearized context of Section V,C, transients certainly decay as I t - CO; it is not known whether the same is , true when the interaction term G is retained. The question may be rephrased ( as follows: suppose U& t ) is a stationary random function of x and t, and b(x, 0) a stationary random function of x, both with zero mean, but with (U A b) # 0 at t = 0, and let b(x, t ) be determined by the exact induction equation with 3, # 0. Does (U A b) tend to zero or to infinity, or to something between as t - CO ? [The related question of what happens to (b) as t - CO , , is an old one (Batchelor, 1950), which has been studied closely from many points of view (Schluter and Biermann, 1950; Moffatt, 1961, 1963; Saffman, 1963; Kraichnan and Nagarajan, 1967) but to which no clear-cut answer has yet emerged.]
.
,
-_
149
The expression (5.35) hears a close relationship with the expression obtained by Parker (1970b) on the basis of his cyclonic events model, namely, aij = 4mirnk( X j ( a ) aXrn/aakd3a ,
(5.37)
where X ( a ) is the displacement of the particle that starts at position a in an event, the angular brackets represent an average over events, and n is the number of events per unit volume per unit time. Expression (5.35) is preferable to (5.37) simply because (5.35) is defined (and therefore meaningful) for any turbulent velocity field, whereas (5.37) is defined only for velocity fields that can be represented in terms of random cyclonic events; although such a representation is physically illuminating, it is unlikely that an arbitrary turbulent velocity field will in general admit such a representation. In the isotropic ( no preferred direction ) situation, (5.35) becomes aij = a a,, where u(t) =
-f
It(.(., t ) * V ,
0
v(a, z)) dz,
(5.38)
and again the appearance of a helicity-type correlation (this time in terms of Lagrangian variables) is to be noted. It would be of great interest to be able to reexpress (5.38) in terms of standard Eulerian correlations (and undoubtedly an infinite series of these would be needed)-but the relationship between Lagrangian and Eulerian statistical quantities presents one of the notoriously difficult central unsolved problems of the theory of turbulence. has The tensor pijk(t) an integral representation similar to (5.35) (Moffatt, 1974). In the isotropic situation this reduces to the form p i j k = b(t)&ijk, where p ( t ) = ~ ( t-)i
j J {<.(a7 t )
0 0 0
Vav(a, 71)
~ ( 8 7 ~
2 ) )
- <v(a, t ) * v(a, z2)Va v(a, zl))) dzl d z ,
+ &(t) 5 a(z) dz,
(5.39)
where D ( t ) = $Iii, and D , and a(t) are given by (5.36) and (5.38), respectively. The last term of this expression ensures that if a ( t )tends to a constant value as t + CO, then on this diffusionless theory p ( t ) is unbounded as t + co.t
t Footnote added in proof: The functions a ( t ) and P ( t ) given by (5.38) and (5.39) have recently been computed by R. H. Kraichnan (private communication) for gaussian turbulence of maximal helicity; the results indicate that both a(t) and P ( t ) tend to constant nonzero values of order uo and uo I, respectively.This means that the divergence in the final term of (5.39) is compensated by an equal divergence in the double integral term. This interesting result may be peculiar to the assumed gaussian statistics.
150
H . K . Moflatt
Hence it is essential to retain the effects of molecular diffusivity in order to determine the appropriate form of fi in the weak diffusion limit when the turbulence lacks reflectional symmetry. This calculation (if it could be done) would inevitably lead to an expression of the form B = u,lf(R,), where f(R,) + CO as R, + a. may be noted that situations such as this are not (It unprecedented: in the problem of longitudinal diffusion in turbulent pipe flow, for example, the eddy diffusion coefficient is inversely proportional to the molecular diffusivity.)
E. THEFORMS aij AND OF
Bijk
IN
AXISYMMETRIC TURBULENCE
Suppose now that the turbulence (or random wave field) is not invariant under all rotations, but only under those about the direction defined by the unit vector e. This situation may well arise when the fluid is in a state of mean rotation R and when the turbulence is in consequence influenced by Coriolis forces [a situation recently realized and studied in the laboratory by Ibbetson and Tritton (1975)l; in this case, e = R/Q and the spectrum tensor Oij(k, w ) is axisymmetric about the direction e. The most general form that Oij can take in such circumstances,consistent with Hermitian symmetry and the incompressibility condition ki Oij(k, w ) = 0, is Oij(k, w ) = 41(kz6ij- k i k j )
+ &(keiej + kpz6ij - p k i e j - pkjei)
+ + 4Jk + q5Jk
where
i+,&ijkkk
A
+ i44Eijkek
e)ikj + 4 f ( k A e)j k, e)iej+ 4g(k A e ) j e i ,
ic$4
+ kz45 + pk&
= 0.
(5.41)
Here, 41,..., 4 6 are functions of k, k * e = kp, and w, 41,. . ., 44being real, and 45 and 4 6 complex. The terms involving dl and 4z(which are scalar functions) are reflectionally symmetric, while those involving &, . . ., 4 6 (pseudoscalar functions) are not. The energy spectrum function E(k, p, w ) and helicity spectrum function F(k, p, w ) are given by
-. .
E(k, H
0) = 2d?aii
= 4nk441
+ k(1 + p 2 ) & ,
+(/Az -
(5.42)
1) Im
46).
F(k, U a)= -47Cik2k,,,&klrn@kl ,, = 8xk4{43
(5.43)
The situation is evidently much more complex than in the isotropic case!
151
The pseudotensors aij and B i j k may still, however, be calculated from formulas (5.27) and (5.28), and symmetry considerations imply that these must now take the form
aij = al 6 ,
Bijk
+ a2eiej,
+ 82Ejkrnemei + 83&kimemej
f B4&ijmemek
9
(5.44)
= Bl&ijk
+ 8 S 6 j k e i + Bsdkiej + f i 7 6 i j e k + Bseiejek
(5.45)
where ai,a 2 , B5, 8 6 , p7, and B8 (being pseudoscalars) may be expressed as weighted integrals of the functions 43, ..., 46 and their derivatives with respect to k and k e, while /I1,B2, f13, and p4may be expressed similarly in and &. The details of the calculation are not of great interest; it terms of 41 is enough to note that in general pi, &, B3, and B4 have the same order of magnitude, as do fis, BS, /I7, B8. and The relation (5.3) between d and Bo (consideringonly the first two terms) then reduces to the form
d = a,B,
+ a2(e B,)B,
- B1 V A B, - B2 e (V A B,)e
- B,(e
V)(e A B,)
+ 8 6 V(B, - e) + B7(e
+ B3(e A V)(B, e) V)B, + &,(e V)(Bo e)e.

*
(5.46)
The a effect, represented by the first two terms, is now anisotropic but has not suffered any fundamental modification. A nonisotropic a effect of this form still gives rise to amplification of field modes proportional to exp iK x provided (Moffatt, 1970a)
a:K
+ alccz(K2- (K
e)) > A4K4.
(5.47)
.p2,
.*
Similarly, the eddy diffusivityeffect is now anisotropic: the terms involving /I3, and p4 (which occur even in reflectionally symmetric conditions) are all (presumably)of a dissipative nature like the term involving B1. The term involving fl6 is of little interest (unless fl6 is allowed to vary slowly as it would in an inhomogeneous field of turbulence) since only the curl of d contributes to (5.1). Similarly the term involving B7 may be replaced by -8,e A (V A B,) (the difference being irrotational). This effect was discovered by Radler (1969a) and was expressed in the form ylS2 A J, , where y1 = -B7po/i2, and is now generally known as the S2 A J effect. The coefficient y1 is a pure scalar, and it is tempting to conclude (as did Radler) that the S2 A J effect can therefore occur when the random motion is reflectionally symmetric; this conclusion does not, however, appear to be justified, since if Oij(k,o)is reflectionally symmetric, then B7 (and so yl) certainly vanishes. This point is of more than academic interest, because it means that if the S2 A Jeffect is present, then in general an anisotropic a effect
152
H . K . Moflatt
will be present also. Incorporation of the S2 A J effect in a model and simultaneous exclusion of the a effect (Radler, 1969b; P. H. Roberts, 1972, Sect. 6) is therefore perhaps somewhat artificial. Nevertheless, it must be said that these models do show that the S2 A J effect, in conjunction with differential rotation can provide dynamo action, the reason being that the electromotive force y,R A Jo tends to generate toroidal current from poloidal and vice versa; the effect therefore provides a possible means whereby the dynamo cycle may be closed. The term involving fi8 was likewise expressed by Radler in the form
Y Z P * V)(Bo * QP, (5.48) and he argued that if the rotation S2, which gives rise to all the nonaxisymmetric contributions to aij and & j k , is weak, then the term (5.48) being cubic in S2 should be negligible. This conclusion must also be questioned; as mentioned above, the coefficients B, and B8 are in general of the same order of magnitude, and it would therefore seem inconsistent to include Radlers S2 A J effect in any model without including also the term of (5.46) involving & .
F. DYNAMO EQUATIONS AXISYMMETRIC FIELDS FOR MEAN INCLUDING MEANFLOWEFFECTS

When there is a mean velocity U(x) in addition to the random fluctuations t), it is no longer realistic to think in terms of homogeneous turbulence; the statistical properties of the turbulence, and in particular the pseudotensors c l i j , & k , will now themselves be functions of position varying on the same scale as the scale of variation of U(x). The equation for the mean field, which we shall now simply denote B, becomes
U% (
dB/dt = V A ( U A B )+ V
A +~ AVz&
(5.49)
where C is still given by (5.3), in which for the moment we simply assume that we know the form of aij(x) and fiijl(x). When B, U, and C are axisymmetnc, it is convenient to separate the toroidal and poloidal ingredients of (5.49). If
B=B&+&,
U=U&+U,,
(5.50)
where U = so(s, z ) in cylindrical polars (s, 4, z), then the toroidal ingredient of (5.49) [cf. (3.8), but note that we have now dropped the subscript 4 from BJ is
aB/at + U, VB = S ( B ~ V)O + (v A &)& + 1(v2- s - ~ ) B . (5.51)

*
153
A
The poloidal ingredient of (5.49)may be uncurled, so that if Bp = V the equation for A [cf. (4.1), but with A = x/s] is
aA/at + U,
*
(Ai,)
VA
= 8,
+ A(V2 - s-2)A.
(5.52)
Note that the 4 component of electric field E is given by E, = -a&%; there can be no electrostatic contribution under axisymmetric conditions. Equations (5.51) and (5.52) make it clear that the meridional velocity Up merely has a redistributive effect with regard to both fields A and B; in the former case, this leads to expulsion of poloidal flux by the mechanism of Section III,B if A is small, although now of course this flux may be regen, erated by the 8 term. The term s(Bp V ) o represents generation of toroidal field from poloidal field by the differential rotation mechanism (Section 111,D).This effect now acts in conjunction with the source term (V A B), ,and dynamos are of two types, depending on which of these terms is of dominant importance. In the simplest situation in which
8 = aB - BVAB,
(5.53)
the B term leads, as we have seen, to an augmentation of the molecular diffusivity 1, the effective diffusivity being simply A, = B + A. The a term gives a contribution V A (aB,) in (5.51). The relative magnitude of the two production terms in (5.51) is then given by
(5.54)
where wb and a. are typical values of I V o and 1 a [ , respectively (such estimates can of course break down locally). If L20b 4 ao, then the differential rotation effect is negligible; the U effect is then solely responsible both for the generation of poloidal field from toroidal and vice versa; dynamos that operate in this way are described as a2 dynamos. The very simple dynamo described at the end of Section V,A is essentially an a2 dynamo, depending as it does on a two-fold operation of the a effect. If L2wb % ao, on the other hand, then differential rotation is responsible for the generation of toroidal field from poloidal, whereas the a effect, through the term U B in (5.52) regenerates poloidal from toroidal field; a dynamo operating in this way is described as an a o dynamo ( a V o would perhaps be more accurate, since it is only the gradient of o that is relevant here). The equations for the a o dynamo in the form
+ U, - VB = SB, .vw + ~ ( v - S - Z ) B , 2 aA/at + Up VA = aB + 1(V2 - s-)A,

aBpt
*
(5.55)
(5.56)
were obtained originally by Parker (1955a).In this form, the sole effect of the
154
H.K. Moffatt
background random motions is the appearance of the term aB in (5.56), and Parker derived this on the basis of his random cyclonic events model (see also Parker, 1970b, 197la,b,c,d,e,f, 1975; Lerche, 197la,b).Equations having the same structure were derived by systematic perturbation procedures by Braginskii (1964a,b), whose approach will be described in the following section; this latter approach leads, moreover, to the appropriate determination of a in terms of properties of the background fluctuating (nonaxisymmetric) motions. The approach that we have adopted in this section (following Steenbeck, Krause, and Radler) seems the most general and the most easily comprehended-and ease of comprehension is of crucial importance when attention is turned to the far more difficult dynamic aspects of the problem. It is, of course, reassuring that the various approaches, from rather different standpoints, do converge at the same destination [in the form of Eqs. (5.55) and (5.56)], and this lends confidence to the extensive analytical and numerical studies of these equations that have been carried out, some of which will be described in Section VII.
VI. Braginskiis Theory of Nearly Axisymmetric Fields

A. LAGRANGIAN TRANSFORMATION INDUCTION OF THE EQUATION
An approach to the dynamo problem that is in some respects complementary to that described in Section V was developed by Braginskii (196k,b) and has since been elucidated and extended by Soward (1972).The approach is based on the idea that, although an axisymmetric field cannot be maintained by axisymmetric motions, weak departures from axisymmetry in both velocity and magnetic fields may be sufficient when I is small to provide the means of regeneration of the field against ohmic decay. In describing the essence of this theory we follow the approach advocated by Soward. The starting point is a property of invariance of the induction equation in the frozen field (I= 0) limit. For reasons that emerge, it is helpful to modify the notation slightly: let B(ji t), ii(ji, t) represent field and velocity at (ji, t), and consider a 1-1 continuous mapping fi x(ji t) satisfying the incomI ( be unity. Then if we pressibility condition that the determinant define new fields
U( , % t ) = ax,/at B,(x, t ) = Bi ax,/aj;., , the invariance property is that the equation
+ iii ax,/agi,
(6.1) (6.2)
aB/at = V ~ ( i i h B )
155
transforms into aB/at = v A (U A B ) . B(x, t ) is (physically) the field that would result from 8(&t ) under the frozen-field distortion I -,x . Suppose now that we subject the full induction equation
(6.4) to the same transformation. We obtain, after some elaborate manipulation of the right-hand side,
aB/at - V A ( U A B )= V A 8 + AVB, where and
a.. = Ae. tj rkl
a8/at - 0A (ii A 8 ) = nPzB
%A(3), aj;-Paxj azP
f l p j = A($& a x . - S d ) . ax
(6.7)
Note that aij (a pseudotensor) and f l p j (a tensor) are both quadratic in the displacement x - W. The similarity between (6.5), (6.6) and (5.1), (5.3) is immediately striking; note, however, that in the present context, the term V A 8 is wholly diffusive in origin. Paradoxically, although diffusion is responsible for the natural tendency of a field to decay, it can also be of crucial importance in creating the electromotive force that can counteract this decay.
B. NEARLY AXISYMMETRIC SYSTEMS

Throughout this section, we shall use angular brackets ( ) to denote an average over the azimuth angle 4 in cylindrical polar coordinates (s, 4, z). Thus for any scalar $(s, 4, z),
.
c
($) = ( 1 / W
j
0
Zn
$b, 4, 2) 4 4
(6.8)
and for any vector f =f,is f b + f,i, , we define 4

( f ) = (f,>is
+ <f4>b+ <f,>i,.
(6.9)
With this convention, we may talk of the mean toroidal field (f4)ig and the mean poloidal field (f,)i, + (f,)i,, which are by definition axisymmetric.
156
H. K . Mofatt
If the velocity field ii(x, t ) is nearly axisymmetric, then we may express it in the form
ii = (ii)
+ &U',
6 1.
(6.10)
*
A magnetic field B(x, t) convected and distorted by such a velocity field must exhibit at least a similar degree of asymmetry, and it is consistent to suppose that
B = (B)
+Eb.
(6.11)
When E = 0, the field (B) cannot survive, by Cowling's theorem. Braginskii's model is based on the expectation that the mean electromotive force & ' ( U A b) may compensate the erosive effects of the term AV2@) in the averaged induction equation. For this to be possible it is necessary that I be not greater than O(E');we therefore put I = I , E' and keep I, fixed as E 0. Expansion in powers of E may then be expressed equivalently as expansion in powers of R, ' I 2 where R, = U , L/A, where U , and Lare scales characteristic of the mean velocity field (ii). This was the procedure adopted by Braginskii. The mean velocity may be expressed as the sum of toroidal and poloidal parts:
(ii) = U(s, z)i,
+ Up(s,
2).
(6.12)
The toroidal part generates toroidal field from poloidal field by differential rotation (Section II1,D) and this is certainly conducive to dynamo action (although, of course, not sufficient in itself). The poloidal part Up(s,2) tends to redistribute the mean toroidal field; it also tends to wind up the mean poloidal field and to expel it (Section II1,B) from regions of closed streamlines of Up (i.e., from the whole fluid region for an enclosed flow) if a magnetic Reynolds number based on a typical value of I UpI is large. This latter process is certainly not conducive to dynamo action, and the only way to control it is to suppose that I UpI/U, is at most O(R;'), so that U , L/I = R, ,
I Up IL/I = O(1).
(6.13)
The dominant ingredient of the mean velocity field is then the toroidal ingredient Ui, whose typical order of magnitude is U,. To emphasize this scaling, we rewrite (6.12) in the form
(C) = U(s, 2 ) t
+ &'U,(S,
2).
(6.14)
Note that, in proceeding in this way, attention is automatically restricted to velocity fields that have some a priori chance of success as potential dynamos.
157
The mean field (B) may likewise be expressed as the sum of toroidal and poloidal parts:
(B) = B(s, z)i,
+ Bp(s, z).
(6.15)
The dominant differential rotation Ui, generates Bib from Bp and it may be anticipated (Section II1,D) that B = O(R,) I Bp 1, or equivalently 1 Bp I = O(E)B.Hence we may also rewrite (6.15) in the form
(B)
= B(s, z)ib
+ E2bp(s,z).
(6.16)
It is, of course, implicit in the notation that B, 1 b 1, and 1 bp 1 are all of the same order of magnitude as E -P 0; similarly for U, I U 1, and I up 1. C. NEARLY RECTILINEAR FLOWS; EFFECTIVE FIELDS It is mathematically simpler to focus attention on the Cartesian analog of the situation described in Section VI,B, in which (s, 4, z ) are replaced by (x,y, z), and ( ) in consequence now indicates an average with respect to the y variable. The additional difficulties of dealing with the cylindrical coordinate system are purely geometrical and need not concern us here. suggests The fact that the fluctuating fields of (6.10) and (6.11) are O ( E ) that it may be possible to accommodate them through the use ofa transformation function
fi = x
+ q ( x , t),
~ ( xz)i, ,
(q) = 0,
q = 0.
(6.17)
If this is possible, then the related fields U and B will have the simple form (the time dependence being for the moment understood)
+ E~U,(X, z) + O(c3), B = B(x, z)i, + EZbep(x, ) + O ( E ~ ) , z

U =
(6.18) (6.19)
where the effective fields U, and be, are related in some way (to be determined) to the fields up and b, . Let us first obtain the relationship between the fields bepand b, . First, by expanding b i ( x ) in Taylor series about the point fi and using hi(%) = (8ik E aqi/ax,)Bk(x), we obtain
B i ( x )= Bi(x) E[VA (q A B(x))],
0(&3).
(6.20)
158
H . K . Moflatt
Comparison with (6.11) shows that
=V
A ( ~ A B ) B aq/ay - (q * V)Bi, =
(6.21)
(to leading order), and the mean poloidal ingredient of (6.20) gives
bp
= be,
+ tv A
( t l ~ aqM/aY)B, A
(6.22)
where qM= q - (q iy)iyis the meridional projection of q. In terms of the vector potentials a, a , defined by
b, = V
(aiy),
be, = V
(a, i,,),
this result takes the simpler form
a = a , - wB,
w = -I < q M 2
aflM/aY)y
(6.24)
The relationship between U and up is similarly obtained with the sole , , difference that it is U k - axk/at (rather than U k ) that appears in the transformation relationship (6.1). This leads to the expression [analogous to (6.21)]
U = (a/at
+ U a/ay)tl - (q - V ) u i , ,
A
(6.25)
and the result [analogous to 6.22)]

up = U , ,
+ +v
(1 1,
(alat
+ U a/ay)q,).
(6.26)
D. DYNAMO EQUATIONS NEARLY FOR RECTILINEAR FLOWS Equations for the evolution of B(x, z, t ) and bep(x,z, t ) may now be obtained by averaging (6.5)with respect to y. The y component (i.e., the toroidal ingredient) of the resulting equation is
swat + A,, - V B = &be, - vu + (v A go)y+ W B +

where
boi = A(Uij)Bj
0(&4),
(6.27)
hikp(Bpj) aBk/aXj .
(6.28)
Since 1 = O(E) and aij and Bpj are both O(E),6, is evidently O(c4), so that to leading order (6.27) becomes
aB/at
+ EzU,, - V B = &be, - vu + I,.WB.
(6.29)
In this equation the effect of departures from exact rectilinearity are wholly , , absorbed through the introduction of the effective fields U and be,. Note that the time scale of evolution of the B field is O ( E - L / U ~ ) .
159
The x and z components of the averaged equation (6.5) may be "uncurled" to give
e2 aa,/at
+ c4uep .Va, = g o y+ c2AV2ae+ O(e5),
(6.30)
and here it is clear that the term goy of the same order as the other terms in is the equation. Moreover, the dominant contribution to do, [from (6.28)] is evidently given by 80, J-(a22)B = where, from (6.7) and (6.17),
(a22>
+ O(e6),
= 2E 2 ( q 3 , m q l , m 2 ) ,
(6.31)
= E2E2kl(?k,
mq1, m 2 )
(6.32)
where a subscript after a comma indicates differentiation, and (6.30) now becomes
aa,/at
+ c2uep - Va, = X B + IZV'a,,
(6.33)
where r = Equations (6.29) and (6.33) (with the correspondences c2ueP- U,, , &'a, - A, c2bep- Bp, A(a22) -+a) are the Cartesian analogs of the a o , , dynamo Eqs. (5.55) and (5.56). The effective value of a, given by
a = 2Je2(r/3, m q 1 , m 2 ) 9
(6.34)
where q is related to the velocity field by (6.25) is noteworthy. The frozenfield displacement merely gives a configuration of u' and b that is potentially able to provide an a effect; it is the diffusive slip of b relative to U that ' realizes this potential and provides an a proportional to 1(cf. Section V,C). The structure of (6.34) is elucidated by considering a particular displace-
(6.35)
for which V A q = kq (so that the displacement has positive helicity) and for which (6.34) reduces to
-1e2k3q$.
(6.36)
So again, as in the random wave approach of Section V,C, positive helicity is associated with a negative value for a. More generally, as pointed out by Braginskii (1964a), a displacement of the form
q = q(")(x, cos ky z)
+ q(')(x,z) sin ky
(6.37)
will provide a nonzero value for a [given by (6.34)only if the vectors q' and ()
160 the form

W
H.K. Moflatt
q(') are linearly independent]. Similarly the parameter w given by (6.24) has
=
- k(q("' A q")),,
(6.38)
when q is given by (6.37), and this also is nonzero only when q(c) q(') are and linearly independent vectors. If terms of order higher than E' are retained in (6.29) and (6.33), then new effects are to be expected. These effects turn out, however, to be very much what might be anticipated from the background of mean field electrodynamics as expounded in Section V. To order .s3,the form of Eqs. (6.29) and (6.33) is in fact unchanged (Tough, 1967; Tough and Gibson, 1969), but the expressions for w and r are modified; in particular, r takes the form To + Erl,where now To is the zeroth-order expression (1/E2)awith a given by (6.34), and rl involves the mean of an expression cubic in q-compare the form of expansion (5.24), which may also be regarded as an expansion in powers of the amplitude of the fluctuating velocity. At order e4 (Soward, 1972) effects analogous to the nonisotropic c1 and /? effects of Section V,E also appear.
E. COMMENTS ON
THE GENERAL APPROACHOF
SOWARD?
Soward's approach undoubtedly provides the most natural framework for the treatment of nearly axisymmetric systems in the limit 1 -,0. However, Soward claims that his approach also provides the natural framework even when departures from axisymmetry are not small [i.e., when the transformation function %(x, t) is not restricted by an assumption of the form (6.17)]. This claim, although plausible, requires some qualification. The reason is that the approach will work, as Soward acknowledges, only for velocity fields ii(S t) for which a transformation f = %(x, t) exists such that the related field u(x, t) given by (6.la) is predominantly azimuthal, in the sense that
u(x, t ) = U(s, z, t)i+
+ O(Ril).
(6.39)
Now it can easily be shown that the streamlines of the field ii(% t) map into streamlines of u(x, t), which according to (6.39) are nearly the circles CO: = so, z = z , ; more precisely the topological measure analogous to s (2.18) is (6.4)
-
t Soward (1972).
161
where U = V A g and U * n = 0 on the surface of and for a field of the form (6.39), K = O(R; '). This means that unless the value of K for the field i 2 t ) (, is O(R; ') (for every volume P with surface s" on which n * i = 0), the transformation function 2(x, t ) simply does not exist. In simpler terms, if the streamlines of i(%) are linked, they cannot be unlinked by a continuous mapping 2 + x. In general, a field having both toroidal and poloidal ingredients will have streamlines that are both knotted and linked (Moffatt, 1969) and so the restriction is in fact a severe one. The criticism is not so serious when weak departures from axisymmetry are considered. In this situation Soward (1971) has proved that, given a velocity field of the form
i = U(s, z)ib + EU'(S, 4, z),
(6.41)
there exists an axisymmetric V(s, z) such that i E'V has streamlines that are closed at the O ( E )level [i.e., if a streamline is followed as 4 increases by 2% the separation of the endpoints is o(E)].It seems likely that this same choice of V [which is identifiable with uep - up in (6.26)] also guarantees that these same closed streamlines must be unlinked, although a convincing proof of this has yet to be given. A mapping 5 -+ x that maps the streamlines of i + e'V onto the circles COthen exists, and the method can proceed with the necessary adjustments at order E' in the meridional velocity represented by ~ , p up + V. Unfortunately there is no analog of Soward's closed streamline theorem when E is not small; hence the limitation in the class of velocity fields to which the method may be applied when large departures from axisymmetry are considered.
F. COMPARISON BETWEEN THE TWO-SCALE NEARLY AND AXISYMMETRIC APPROACHES
The many points of contact between Braginskii's theory and the theory of Section V are too numerous to be merely fortuitous and lead to the view that the two approaches are in fact different aspects of one and the same theory. What is essential in both is that it should be possible to define an average ( ). The existence of two scales, 1 and L, in Section V, lead to the conclusion that d = (U A b) depends only on the local values of B, and its space derivatives and of the statistical properties of the velocity field. In Braginskii's theory, the two scales are also present in disguised form, the small scale being simply the distance that a fluid particle migrates from its mean position during its passage around the axis of symmetry. The same considerations as in mean field electrodynamics lead inexorably to the conclusion that
162
H. K. Mofut
8 must depend only on local values of B(s, z), U(s, z) (and possibly their derivatives), and the mean properties of the fluctuating motion. The crucial conclusion of both theories is that a contribution to 6 of the form a ( B ) can arise; and since a is a pseudoscalar in either theory, this effect necessarily requires a lack of reflectional symmetry in the motions that give rise to it; i.e., an observer would be able to distinguish (locally) either a right- or a left-handed character in the motions if the resulting value of a is nonzero. The distinction between uep and up (and between bePand bp) must be regarded as merely of academic interest until a complete dynamic theory accounting for the structure both of the up field and of second-order mean quantities [like (6.34)] is available; in any case, if q vanishes at the boundary S of the region V of conducting fluid, then bp = beP S, and so the matching on conditions with the external field are unaffected by the distinction between b, and bep. One is therefore bound to ask what advantage Braginskii's theory has over what appears to be the simpler and more illuminating approach of Section V. The answer is simply that it provides the means whereby the pseudotensors aij and Bijk may be derived in the important situation when the mean flow is predominantly toroidal and axisymmetric, the fluctuating velocities are weak, and U,L/L is large. The theory may therefore be recognized as falling within the general framework of the mean field electrodynamics of Section V, but at the same time as occupying a particularly difficult and thorny corner of that subject. It is nevertheless necessary to grasp the thorns, since the most plausible models for both Earth and Sun do invoke strong toroidal velocity fields and one can hardly simply ignore the influence that this may have on aij and f l i j k , particularly when dynamical effects are subsequently considered. Perhaps a more obvious attractive feature of Braginskii's approach is the very degree to which it places emphasis on the nonaxisymmetric ingredient ~b of magnetic field. The Earth's field is of course nonaxisymmetric relative to the axis of rotation, the axis of the magnetic dipole moment being inclined at a slowly varying angle 1+9(at present of order 10")from the rotation axis. A perturbation field having a dependence exp i(4 - ot) on azimuth 4 about the rotation axis (Braginskii, 1964c) gives a field with external potential of the form
0
.
-
A~-zP;(COS cos(4 - ot), e)
(6.42)
i.e., a dipole whose moment rotates in the equatorial plane with angular velocity o.In conjunction with the axial dipole, there is here the basis of one (though certainly not the only) possible explanation for the tilt and the manner in which the direction of the dipole moment vector drifts relative to the rotation vector.
+,
163
VII. Analytical and Numerical Solutions of the Dynamo Equations
A. THEa2 DYNAMO a CONSTANT WITH

We have already discussed in Section V,A the simplest exponentially growing solutions of the equation aB/at = V A (U A B )
'
'
+ V A (aB)+ 1, V2B,
(74
in the case when U = 0, a = const, and the fluid is of infinite extent, namely the force-freemode for which V A B = KB, for which the growth rate is given , Ione , by (5.9). Now that we have estimates of a and 1 (= 1 + / ) further comment on this solution is called for. The growth rate o is clearly maximal when K = K , = a/21,. When R, = U, l/A -4 1 (in the notation of Section V), a lui/1 and A, 1,so that
lK,
R i 4 1,
and so the initial assumption that the scale L of the mean field ( K ; ') be large compared with the scale 1 of the random U field is amply satisfied. At the other extreme, when R, % 1, we have seen that there is some uncertainty in estimates of a and p owing to uncertainty regarding convergence of the integral (5.39), and the unbounded behavior of p as given by (5.40). If we assume that a is independent of A in the limit 1 -,0, then on lu, 1 , If 1 is also 1. , dimensional grounds, a uo in this limit, and so lK, independent of 1in the limit 1+ 0 (as maintained, e.g., by Parker, 1971b) , then again on dimensional grounds 1 uo 1. This would lead to the conclusion that IK, = O(l), i.e., that the medium is unstable to magnetic disturbances on scales right down to the scale 1 of the turbulence itself. Of course, if K1= 0(1),then the two-scale approach ceases to be valid; nevertheless the increase in growth rate o with decreasing scale K-' calls for considerable delicacy in the interpretation of results in this limit. This particular difficulty is removed if, as argued at the end of Section V,D, fi is in fact large compared with lu, when R, % 1. The inadequacy of the traditional estimate p lu, in the solar context has been emphasized by Piddington (1972a); however, Piddington's main concern is with the process by which small-scale field fluctuations merge and annihilate each other, and the relevant eddy diffusivity for this process is not necessarily the same as the eddy diffusivity that operates on the global mean field. In this context, see also Parker (1972, 1973). The a2 dynamo in a spherical geometry was treated by Krause and Steenbeck (1967).This requires solution of (7.1) (again with U = 0, a = const) in a
(74
164
H. K. Mofatt
sphere of radius R, and with matching to a current-free field for r > R. The solutions in this case cannot be force-free, since as mentioned in Section 11,D such fields do not exist in a finite geometry in the absence of external sources; the solutions nevertheless undoubtedly have a helical structure. Steady solutions are possible only if the parameter aR/& takes one of a discrete set of eigenvalues; the lowest of these is 4.49 and the corresponding solution (in r < R ) for A and B has the form A = C{k-'r-'12J3,2(kr) + - J l , 2 ( k r )sin 8, }
B = Cr-'/2J3/2(kr) 8, sin
(7.3)
\ I
which matches to a dipole field for r > R. If aR/1, $ 1 (as would, for exam, ple, be the case if a - U o , 1 uo 1, and 1 -g R) then a wide range of modes become unstable, and the most unstable is not that having the simplest structure (7.3) but rather a high harmonic having a much smaller length scale (as suggested by the results for an infinite geometry).
B. a2 DYNAMOS ANTISYMMETRIC WITH a

The assumption of constant a is unrealistic in a rotating body such as the Earth or the Sun in which Coriolis forces are responsible for the generation of helicity. We referred in the introductory section to Parker's (1955a) concept of rising, twisting blobs of fluid in the convection zone of the Sun surrounded by connected regions of falling fluid. The rising blob tends to entrain fluid from the sides and conservation of angular momentum then leads to spin-up or positive helicity in the northern hemisphere, and similarly to negative helicity in the southern hemisphere. Correspondingly, a would in this picture be negative in the northern hemisphere and positive in the southern. [It is relevant to note that Steenbeck et al. (1966) came to precisely the opposite conclusion in arguing that a blob that rises through several scale heights will expand and therefore "spin down" in the northern hemisphere, generating negative helicity there. Viscous entrainment and compressibility effects are in competition here and the true sign of a-which is of crucial importance for a o dynamos discussed below-is presumably determined by which of these effect dominates.] The simplest reasonable assumption for U ( X ) on this type of physical picture is
a(.)
= a , f ( r / R ) cos
8.
(7.4)
Unfortunately, even with this simplest choice, the eigenvalue problem
VA(UB) 1,V2B = 0 +
(7.5)
165
(with the matching conditions to a current-free field on r = R) is not amenable to simple analysis, and the problem must be solved numerically (Steenbeck and Krause, 1966, 1969b; P. H. Roberts, 1972), using series expansions for B and truncating after a few terms. Roberts found that six terms were in fact sufficient to give 0.1 % accuracy in the determination of eigenvalues. The most striking feature of the numerical results is that the eigenvalues for fields of dipole and quadrupole symmetry are almost indistinguishable, i.e., these fields can be excited with almost equal ease! The reason for this was given by Steenbeck and Krause (1969b): the a effect given by (7.4) operates most effectively near the poles, where cos 8 = & 1, and the toroidal current distribution (which gives rise to the poloidal field) is correspondingly concentrated in rings near the poles. The mutual inductance between these current rings is small, and so one of the current rings can be reversed without greatly affecting conditions near the other. This operation transforms a field of dipole symmetry into one of quadrupole symmetry or vice versa.
c. LOCAL BEHAVIOR aw DYNAMOS IN

4
The Cartesian analog of (5.55) and (5.56) [cf. (6.29) and (6.33)] is aAlat
(7.7) where the variables are functions only of x and z. It is illuminating (Parker, 1955a) to consider the local modes of behavior of solutions of these equations in regions where U , , a, and V U may all be regarded as constant. In interpreting these solutions in the solar context, the axis Ox is south, Oy is east, and Oz is vertically upward, relative to some origin 0 in the convective zone in, say, the northern hemisphere. The equations admit solutions proportional to exp(wt iK * x) (where K is a real two-dimensional wave vector in the x-z plane), the dispersion relation being
+ U, aBlat + U ,
- V A= +AV~A, - V B ~p - vu + W B ,
UB
(7.6)
(w
+ 1K + iu,
K) = - ia(K A V U ) , = 2iy,
(7.8)
say. If y < 0, the roots of (7.8) are given by
.
a
w = -1K - iu,
K k I y I1I2(1- i),
(7-9) (7.10)
and there is an exponentially growing solution of K satisfies
C~IKAVU~~>~K~.
This growing mode has the phase factor exp i(K x - 4 Kt - 1 y I1/t).If U = 0 and U = U ( z )only, then this result implies that ifa aU/az < 0 then a ,
166
H . K. Moflatt
growing mode with space dependence exp(iKx) can propagate in the positive x direction, i.e., toward the equatorial plane. (Alternatively, we could regard w as real and K as complex, in which case the growth of the mode is spatial rather than temporal.) Conversely, if a aU/az > 0, then the growing mode propagates away from the equatorial plane. The sign of the product of a and the vertical shear is therefore of crucial importance in determining the character of these local solutions, and this carries over also to properties of global solutions of (7.6) and (7.7) in a spherical geometry (Roberts, 1972). Note that when up = 0, the above solutions are necessarily oscillatory; the and we phase velocity can, however, be reduced to zero if up * K = - I y then have an exponentially growing field whose spatial structure remains constant. This property again carries over to global a o dynamos (Braginskii, 1964b; P. H. Roberts, 1972). Suitable choice of meridional circulation can convert a situation in which an oscillatory dynamo (as described in Section I1,A) is preferred into one in which a steady dynamo is preferred, i.e., excited D at a lower value of I I [see (7.12)]. If this meridional circulation is too strong, however, the dynamo fails altogether, presumably because of the consequent expulsion of poloidal flux from the region where the differential rotation would otherwise be generating toroidal field from it. Since differential rotation and meridional circulation are dynamically coupled, the question of whether a given system will exhibit steady or oscillatory dynamo behavior inevitably demands consideration of the governing dynamical equations. D. GLOBAL BEHAVIOR am DYNAMOS OF When a is given by (7.4) and o is assumed to vary only radially, say
o(r) = obg(r/R),
(7.11)
the crucial dimensionless parameter is what Parker called the dynamo number
D
= a o o bR4/A2.
0
-
(7.12)
Generally, when f and g are reasonably smooth overlapping functions and up = 0, (5.55) and (5.56) yield (numerically) only oscillatory solutions (P. H. Roberts, 1972),a solution of dipole symmetry being more easily excited than one ofquadrupole symmetry when D < 0. In such oscillatory solutions, there is a tendency during a period of the oscillation for toroidal field to propagate from polar regions toward the equatorial plane (consistent with the discussion of Section VI1,C); it is this type of behavior that lends credence to the picture relating sunspot formation to the behavior of the underlying toroidal field, as outlined in the introduction. Indeed, many authors (e.g., Steenbeck
167
and Krause, 1969a; Leighton, 1969;Roberts and Stix, 1972) have succeeded, on the basis of such solutions, in constructing butterfly diagrams, which bear at least a plausible resemblance to those based on observation (Maunder, 1922; see, e.g., Kiepenheuer, 1953). Such diagrams depict the appearance of sunspots as a function of latitude and time and are extracted from solutions of the dynamo equations by plotting in the 8-t plane the isotors B(r, 13,t ) = const, for some fixed value of r, on the view (Parker, 1955b)that sunspots form by eruption when the toroidal field below the surface exceeds some critical value. When the functions f and g are nonoverlapping, the field most easily excited can be steady rather than oscillating. The case
f ( P ) = 4 P - Pl),
g ( P ) = S(P - P 2 ) ,
P1
P P2
(7.13)
,
b
has been studied by Dienzer et al. (1974), who conclude that the steady field is favored when p1 - p2 1 is sufficiently large. As an alternative to the eigenvalue approach, Jepps (1975) has carried out a step-by-step numerical integration of (5.55) and (5.56) again for particular choices of the function f and g and for varying values of D . Numerical experimentation leads to determination of the critical value of D at which there exist solutions varying periodically in time without long-term growth or decay. An advantage of this approach is that dynamical effects can be readily incorporated. In particular, Jepps investigated the effect of making U a decreasing function of the local value of B (see Section VII1,A); he found that the sinusoidal time behavior of the poloidal field was distorted by this sort of nonlinear effect toward a spiked rather than a flattened wave form. Finally, we note that Parker (1971b)has proposed that Eqs. (7.6) and (7.7) with up = 0 may also provide a correct description of the process of generation of the galactic magnetic field (a view that is strongly contested by Piddington, 1972b). With origin 0 at an arbitrary point of the plane of symmetry of the galactic disk (of thickness 2z0), Oz normal to the plane, Ox radial, and with
u = ~ x , a=(
a0
-U0
o<z<z,, -zo < z < 0,
(7.14)
Parker obtains the dispersion relation for nonoscillatory modes proportional to exp(ot + ikx): for large positive values of the dynamo number, which in this case is D = aGzi/A2, (7.15) and for modes in which B is symmetric about z = 0, this takes the form
w
- - (AD/kz:) exp(
-@I3)
cos 4,/3D/3,
(7.16)
168
H. K. Mofatt
indicating exponential growth if (4n - 1).
-= f i D1l3 < (4n + 1 ) s
(7.17)
where n is any large positive integer. This is a surprising result since it indicates that for large values of D,a relatively small change (positive or negative) in D can transform a situation that permits exponential growth of a magnetic field mode into one in which the same mode decays. Of course, it may be that oscillatory modes are more easily excited (Parker, 1971e),as is the case for the spherical geometry; but if not, and if Parker's model is applicable, then the nature of the relation (7.16) between o and D would suggest that the galactic field depends for its existence on a precarious interaction between effects of horizontal shear and cyclonic turbulence of rather critically defined intensity.
VIII. Dynamic Effects and Self-Equilibration A. WAVES INFLUENCED CORIOLIS BY FORCES, AND ASSOCIATED DYNAMO ACTION A suitable starting point for the consideration of dynamic effects is an analysis of the waves that can propagate in a rotating incompressible fluid permeated by a locally uniform magnetic field Bo. In this context, it is /B convenient to use the local Alfven velocity V = (pop ) - ''O as the measure of the field strength. The linearized equations governing velocity perturbations U about a state of rest and magnetic perturbations b = (p0p)'/'v are then aulat 2 h U = -VP + v vv + v v ~ u ,
avlat = v vu + wV, v U = v v =o,

a
(8.3)
(8.4)
where
p =p/p
+ (v + V)'/2P - (a
a) exp i(k
0
A X)',
and v is the kinematic viscosity of the fluid. These equations (first considered by Lehnert, 1954) admit solutions of the form
(U, v,
.
"
P) = (U, Q,
x - at),
U
(8.5)
and (8.2) gives the important relation between Q and Section V,C: Q = - ( U + iAk')-'(k * V)U.
already implicit in (8.6)

Equation (8.1) then gives
t
169
-io%
+ 2QAu = -ikp,
*
(8.71
where
(8.8 ) By crossing (8.7) with the vector k and then repeating this operation, we obtain two equations from which U and k A 8 may be eliminated, giving the dispersion relation
Q
=w
+ ivk2 - (o+ iAk2)-(V
k).
T 2(k Q)/k,
(8.9)
(8.10)
and the corresponding relation
ikA8 = fku
between the (complex) components of 8. This latter relation implies that such a wave is circularly polarized [cf. thnpatial structure (6.35)] and that its helicity is
(U - 0 ) =
*
f+kIBI.
(8.11)
If, therefore, for any reason waves corresponding to, say, the upper sign are present in greater proportion than those with the lower sign, then the net mean helicity will be positive, and the medium will be unstable to the growth of large-scale field fluctuations, by the a effect discussed in Section V,A. If now we imagine that the field V itself varies on a large length scale, then large-scale Fourier components of force-free structure [like (2.25) but with I K I much smaller than typical values of I k I in the spectrum of U] will grow exponential1y. What then ultimately limits the growth of this type of field structure? To see this, it is only necessary to consider the detailed form of the integral (5.27) for aij, which is, of course, still applicable. As V grows locally in strength, the dispersion relation o = o ( k ) as given by (8.8) and (8.9) is progressively modified in such a way that all components of aij (which is severely anisotropic in this situation) tend to decrease. Field intensification ceases when these components are reduced to such a level that ct effect generation is just compensated by ohmic dissipation. If there is no source of energy for the small-scale motion, then both velocity and magnetic fields must ultimately die away through both ohmic and viscous effects; on the other hand, if random body forces are invoked as a means of maintaining a random wave field, then a steady magnetohydrodynamic equilibrium may be anticipated. This type of process has been analyzed in detail (Moffatt, 1970b, 1972; Soward, 1975). Of particular interest is the level to which the magnetic energy density M = 3V can grow in this situation (the overbar
170
H . K. Mofatt
*
indicating an average over the large wavelength 2 4 K of the V field). The result obtained (Moffatt, 1972) in the case when waves of typical frequency are maintained by a random force distribution is that in the ultimate steady state the ratio of M to the kinetic energy density E in the random wave field is
M/E
(Q/wo)2(v/~)2)(L/1),
(8.12)
where 1 is as usual the scale of the U field and L the scale of the V field. [Viscosity enters into this calculation because, under random forcing, there are some modes (those with k V = 0) that are unaffected by the magnetic field, and it is only viscosity that can limit the growth of such modes when excited at their resonant frequencies.] Clearly M can be large compared with E if L/l is sufficiently large: there is no question ofequipartition ofenergy in a dissipative system of this kind. If we neglect dissipative effects in the dispersion relation (8.8) and (8.9) (i.e., put 1 = v = 0), then when J V* k J 4 4(S2 * k)/k2, the two roots of the resulting quadratic for w are approximately
of 2 0 COS 0 x
+ (2Q COS O)-(V
k ),
O,
z -(2Q
COS
0)(V * k) (8.13)
4
where 8 is the angle betweeri 11dnd k. The root wf corresponds to an inertial , I) wave slightly modified by the Lorentz force. The root 0 ( I w, 4 /of corresponds to a very slow wave (relative to the time scale Q- ) in which , inertia effects are almost negligible. Indeed, the root 0 may be most simply obtained from (8.1)-(8.3) by first dropping the inertia term au/dt. The resulting force balance between Coriolis, Lorentz, and pressure forces is known as magnetostrophic balance [a term introduced by Malkus (1959), although not quite with this same meaning], and we shall therefore describe the waves corresponding to the root w, as magnetostrophic waves. [Hide and Acheson (1973) in their review of the hydromagnetics of rotating fluids use the term hydromagnetic-inertial waves, but insofar as inertia plays no part (except via the Coriolis force), the term is perhaps misleading.]
0
-
B. MAGNETOSTROPHIC AND FLOW
THE
TAYLOR CONSTRAINT
There is good reason to believe (Hide and Roberts, 1961) that the global force balance in the core of the Earth is approximately magnetostrophic, i.e., that to a good approximation,
2 p a A u = -vp
+ JAB,
(8.14)
171
where for the moment we use capital letters U, J, B to emphasize that we have large-scale fields in mind. Suppose that the fluid is contained within a sphere S, center 0, with U n = 0 on S, and let C(so) be the cylinder s = so, where (s, 4, z ) are cylindrical polar coordinates with Oz parallel to Q. Then integration of the 4 component of (8.14) over C(s,) (Taylor, 1963) gives (8.15) the integral being over that part of C(so)contained within S . Under conditions of magnetostrophic balance, therefore, the magnetic field B and assoL ciated current J = pg V A B must satisfy the Taylor constraint (8.15). The result is unaffected by the inclusion of buoyancy forces in (8.14), which of course have no 4 component. We shall refer to important implications of this constraint in Section VII1,D.
c. EXCITATION MAGNETOSTROPHIC WAVES BY OF (MAC)

UNSTABLE STRATIFICATION
One of the great difficulties in making firm progress on the dynamics of the Earths interior is the absence of direct knowledge concerning the dominant source of energy for core motions. One possibility, however, is that at least some part of the liquid core is unstably stratified (either thermally or through coexistence of mixed ingredients of different densities-Braginskii, 1964); the question of thermal stratification has been discussed by Higgins and Kennedy (1971) and Kennedy and Higgins (1973) on the basis of existing knowledge concerning the melting point temperature of iron under high pressure, and the conclusion of the latter study is that the inner one-third of the liquid core may have a temperature distribution that allows adiabatic convection, whereas the outer two-thirds is (in this sense) stably stratified. Whatever the actual situation, the followkg idealized problem (Braginskii, 1964d, 1967, 1970; Eltayeb, 1972, 1975; Roberts and Stewartson, 1974, 1975) provides a basis for careful analysis of the effect of unstable stratification. Suppose that fluid is contained between planes z = f z , on which the temperature 0 is 8, T BZ,, respectively. Suppose further that R = (0, 0, Cl) and V = (0, 0) (other directions also for R and V are considered in the papers of Eltayeb, 1972, 1975); then, on the Boussinesq approximation, Eq. (8.1) is modified only by the inclusion of a buoyancy force age on the right-hand side, where CI is the coefficient of thermal expansion, and 8 satisfies the linearized heat conduction equation
aelat
+ pu, = uv20,
(8.16)
172
H . K . Mofatt
with K the thermal diffusivity. Equations (8.2) and (8.3) continue to hold unaltered. These equations admit solutions proportional to exp[i(Zx + my) + at], with further U,, U,, u x , and U , proportional to cos nz, and u Z , U,, and 0 proportional to sin nz. If viscous effects are neglected and if the planes are assumed perfect conductors of heat and of electric current (conditions that are again varied in the papers by Eltayeb, 1972,1975), then all the boundary conditions are satisfied provided nz, /z is an integer. If &/at is also dropped from the equations (so that attention is focused on the slow magnetostrophic modes) the following cubic equation determining possible values of o is obtained:
Y 2 ( o 1k2)(o
+ K k 2 ) + (U + K k 2 ) - X(W + 1k2) = 0, + Z2)/m2k2V2,
(8.17)
where
k2 = Z2
+ m2 + n2,
X = aPg(m2
Y = 2nJZ/VmZk. (8.18)
In the nondissipative limit 1 = K = 0 to which Braginskii (1964d, 1967) restricted attention, the roots of (8.17) become o = 0 and
0=
+(X
- 1)12/Y,
(8.19)
indicating instability (a2 0) whenever k = (1, m, n) is such that X > 1. In > the geophysical context, as in previous sections, we interpret the y direction as east (or azimuth), so that m (like n ) can take only discrete values. The most unstable mode is that for which m takes its.smallest nonzero value m, (corresponding to exp id dependence on azimuth), and as pointed out by Roberts and Stewartson (1974), if gab/V2 exceeds m: by any amount, no matter how small, an infinite number of modes (corresponding to large values of 1) become unstable, since as 1 -+CO, X aPg/V2m:. Of course, as 1 increases, these modes are increasingly affected by diffusion and also by inertia [neglected in obtaining (8.17)]. Weak diffusion plays a more important role in shifting the root o = 0 away from the origin of the complex plane. Naive linearization of (8.17) in the small quantities 1,K, and o gives for this root the expression
o = -1k2(X - q ) / ( X - l),
= K/A.
(8.20)
-.
If 0 < q < X < 1, then the modes given by (8.19) are stable, but that given by (8.20) is unstable with a slow growth rate determined by the weak diffusion effects. This is a resistive instability in the terminology of Furth et al. (1963); its existence underlines the possible importance of diffusion effects, no matter how weak these may be. It is not clear that the arguments of Higgins and Kennedy (1971) concerning the impossibility of radial convec-
173
fl
tion with no diffusive heat exchange apply with equal force when resistive instabilities of this kind are considered. In general, the three roots of the cubic (8.17) are either all real or one is real and the other two are complex conjugates. In this latter case, if the , complex conjugate roots are given by w = w, iwi with 0 > 0, the corresponding disturbances propagate as magnetostrophic waves of increasing amplitude. Roberts and Stewartson (1974) show the region of the plane of the variables q and Q = 2R1/V2 in which such unstable modes are excited in preference to steady modes for which wi = 0. The distinction between overstable modes and steady modes is of potential importance in the context of the main problem of regeneration of the large-scale field. We have seen in earlier sections that lack of reflectional symmetry, as measured by the mean helicity (U * V A U) is vital in this context; the mean helicity of the unstable motions of the present problem can be calculated on the basis of the linearized equations, and it is found that (U * V A U) (the average being over horizontal planes) is nonzero only if wi # 0, i.e., only if the modes are of the overstable variety. It is possible that the nonlinear effects studied by Roberts and Stewartson (1974, 1975), and in particular the inclusion of mean shear in the y direction, may modify this conclusion, but this is something that requires further investigation; in this context, see also the discussion of Roberts and Soward (1972, p. 148). At any rate, the picture conceived by Braginskii (1967) is the following: unstable stratification in the liquid core of the Earth in the presence of the predominantly toroidal field excites magnetostrophic waves that are essentially nonaxisymmetric and that propagate in the azimuthal direction. Braginskii called these waves MAC waves (M for magnetic, A for Archimedean, C for Coriolis). These nonaxisymmetric motions provide the tl effect that is necessary to generate poloidal field from toroidal field, which in turn is generated from the poloidal field by dominant differential rotation. We have here the possibility of a dynamically consistent dynamo driven by buoyancy forces; to be completely consistent, however, the stability analysis described in this section should include the effects of a mean flow in the y direction with strong vertical shear (i.e., the Cartesian analog of strong differential rotation). Roberts and Stewartson (1975) take an important step in this direction. At the same time, a fully consistent model should include the dynamics of the mean flow itself, an important topic on which we focus attention in the following section. Before leaving the topic of convection-driven dynamo action, however, we should refer to the papers by Childress and Soward (1972), Soward (1974), and Busse (1973), in which different approaches to the determination of a dynamically consistent dynamo are considered. Busse considers a combination of two-dimensional convection rolls, together with superimposed shear
174
H . K . Moffatt
flow parallel to the rolls. Childress and Soward consider convection in a strongly rotating system in which the horizontal scale of the convection cells is small [O(Q-/)] relative to the vertical scale. This permits the use of the two-scale methods described in Section V, and leads (Soward, 1974) to the determination of a dynamo in which the magnetic energy density fluctuates about a weak average value. In these studies involving both magnetic field and rotation, it is well known (Chandrasekhar, 1961) that while rotation and magnetic field are separately stabilizing, the two effects can work against each other in such a way that a flow that is stable under the action of rotation alone becomes unstable when a magnetic field is also introduced, while a flow that is unstable under the action of rotation alone becomes even more unstable with a magnetic field. This consideration led Malkus (1959), on the basis of his appealing conjecture that a convective system with infinitely many degrees of freedom will take advantage of this freedom to maximize the transport of heat in the direction of the imposed temperature gradient, to argue that a magnetic field must grow by dynamo action, if only to release the constraints of rotation, to permit more vigorous convection, and so to increase the transport of heat. Sowards work should in principle provide a test of whether Malkuss conjecture is true or false; but this still remains a choice target for future investigation.
D. MEAN FLOW EQUILIBRATION

A magnetic field that is growing in intensity as a result of the U effect will in general not be force-free and will therefore tend to drive a mean velocity, or to modify any preexisting velocity distribution. This effect has been studied by Childress (1969), Malkus and Proctor (1979, and Proctor (1975). It is supposed in these investigations that background turbulence, present for whatever reason, gives rise to an isotropic U effect, with U a prescribed function of position in a sphere, and that there is initially zero mean velocity. It is further supposed that the level of the a effect is just a little greater than that at which field excitation takes place. The growing field B will then approximate to the axisymmetric eigenfunction corresponding to the lowest eigenvalue for a steady u2 dynamo (Sections VII,A and B). A mean axisymmetric velocity field U(x, t) will then develop according to the equation aulat U vu 2i2 A U = -VP + (p,,p)-1(v A B) A B VVW, (8.21)
and subject to appropriate boundary conditions on the surface of the sphere. The existence of a mean velocity U then modifies the structure of the field B, which evolves according to the equation aB/at = V A ( U A B ) V A ( ~ B+ lV2B ) (8.22)
+
. I
175
in the fluid region. It is to be expected that this modification in structure will lead to increased ohmic dissipation, and hence to curtailment of growth of magnetic energy. This expectation is borne out by numerical integration of (8.21) and (8.22) (Proctor, 1975) in the geophysically relevant case when a = a. cos 6. This problem is characterized by three dimensionlessnumbers, E = v/sZR2, E = I/RR2, M
8 = aoR/I,
(8.23)
where R is the sphere radius, and the lowest eigenvalue of the kinematic a2 dynamo problem (with U = 0) is in this case (P. H. Roberts, 1972)8, = 7.65. E represents the ratio of viscous forces to Coriolis forces, and, since I / L is the relevant velocity scale, E, represents the ratio of inertia forces to Coriolis forces. Both E and EMare small in the geophysical context; in the limit E = 0, EM= 0,(8.21) degenerates to (8.14) and the balance of forces is then magnetostrophic. In each of the cases studied, Proctor found that the growth of magnetic energy M ( t ) was arrested by the mean flow effect, the level of equilibrium M magnetic energy increasing with 8 - 8,. Provided E and E were not too small, M ( t ) settled down to a constant level M(co), after some damped oscillations about this level. Figure 4 shows the structure of the magnetic
FIG.4. Field structures in the Proctor (1975) steady state dynamically equilibrated dynamo; E = 0.01, E , = 0.04, & = 8.0. (a) Poloidal field lines, (b) isotors B = const, (c) lines of constant angular momentum about axis of symmetry, (d) lines of constant vorticity of meridional circulation, (e) streamlines of meridional circulation.
field and the velocity field in the ultimate steady state when E = 0.01, EM= 0.04, and 8 = 8.0. The poloidal field is shown only for r < R, although it of course extends outside the sphere. There is an indication of a cylindrical structure in the patterns of angular momentum and vorticity, which is presumably an indication of the influence of the Taylor constraint (8.15), which is nearly satisfied for these low values of E and EM. E, At even lower values of E and E (E = 0.005, = 0.0025), Proctor found M that the oscillations of M ( t ) about its ultimate mean level did not show any tendency to die out with time. These oscillations of M ( t )are associated with
176
H . K . Mofatt
torsional oscillations about an equilibrium in which the Taylor constraint is satisfied. As pointed out by Roberts and Soward (1972) such torsional oscillations will be damped by Ekman suction effects in boundary layers on the sphere r = R, but when E and E, are very small, this damping is evidently not sufficient to suppress the oscillations. It appears that the steady equilibrium state satisfying the Taylor constraint is unstable or in some other sense unattainable; a related phenomenon was found by Roberts and Stewartson (1975) in the treatment o f nonlinear aspects of the stability problem discussed in Section VII1,C. Neglect of the suppression of the a effect by the growing magnetic field is justified in Proctors investigation for small values of & - 2, since then the mean flow equilibrium mechanism causes a leveling off of M ( t )at a low level at which the turbulence suppression effect is still negligible. The two effects i would have to be considered in conjunction if t - LiCwere large; but then one would also have the complication that more than one mode of the linearized kinematic problem might be unstable. The model is also open to criticism on the grounds that the reflectionally asymmetric turbulence giving rise to the a effect is simply assumed present, without any consideration of the means whereby it may be maintained. A marriage of the models of Sections VII1,C and D may in the longer term provide something approaching the complete picture as far as maintenance of the Earths field is concerned.
,,
ACKNOWLEDGMENTS The writing of this review was completed during my tenure of a Visiting Professorship at the Institut de Mecanique Theorique et Appliquee, Universite Pierre et Marie Curie, Paris, and I wish to express my gratitude to Professor Cabannes and his colleagues for generous support and hospitality during my visit. I am indebted to Dr. R. H.Kraichnan who drew my attention to an erroneous conclusion in the original draft of Section V,C; this led to an important modification of the penultimate paragraph of that section.
REFERENCES
BABCOCK, W., and BABCOCK, D. (1955).The Suns magnetic field. Astrophys. J . 121, H. H. 349-366. BACKUS,G. E. (1958). A class of self-sustaining dissipative spherical dynamos. Ann. Phys. New York 4 312-441. , BATCHELOR, . K. (1950).On the spontaneous magnetic field in a conducting liquid in turbuG lent motion. Proc. Roy. Soc., Ser. A 201, 40-416. BATCHELOR, K. (1953). G. Homogeneous Turbulence. Cambridge Univ. Press, London and New York.
177
S
r
BATCHELOR, K. (1956). Steady laminar flow with closed streamlines at large Reynolds numG. bers. J . Fluid Mech. 1, 177-190. BRAGINSKII, I. (1964a). Self excitation of a magnetic field during the motion of a highly S. conducting fluid. Sou. Phys.-JETP 20, 726-735. BRAGINSKII, I. (1964b). Theory of the hydromagnetic dynamo. Sou. Phys.-JETP 20, S : 1462- 1471. BRAGINSKII, I. (19644. Kinematic models of the Earths hydromagnetic dynamo. Geomag. S. Aeron. 4, 572-583. BRAGINSKII, I. (1964d). Magnetohydrodynamics of the Earths core. Geomag. Aeron. 4, S. 698-712. BRAGINSKII, I. (1967). Magnetic waves in the Earths core. Geomag. Aeron. 7, 851-859. S. BRAGINSKII,. I. (1970). Oscillation spectrum of the hydromagnetic dynamo of the Earth. S Geomag. Aeron. 10, 172-181. E. BULLARD, C. (1968). Reversals of the Earths magnetic field. Phil. Trans. Roy. Soc. London, Ser. A 263,481-524. BULLARD, C., and GELLMAN, (1954). Homogeneous dynamos and terrestrial magnetism. E. H. Phil. Trans. Roy. Soc. London, Ser. A 247, 213-278. BUSSE, F. H. (1973). Generation of magnetic fields by convection. J . Fluid Mech. 57,529-544. BUSSE, H. (1975). A necessary condition for the geodynamo. J. Geophys. Res. 80, 278-280. F. CHANDRASEKHAR, S. (1956). On force-free magnetic fields. Proc. Nat. Acad. Sci. U.S. 42, 1-5. CHANDRASEKHAR,(1961). Hydrodynamic and Hydromagnetic Stability. Oxford Univ. S. Press (Clarendon), London and New York. CHILDRESS, (1969). A class of solutions of the magnetohydrodynamic dynamo problem. In S. The Application of Modern Physics to the Earth and Planetary Interiors ( S . K. Runcorn, ed.), pp. 629-648, Wiley, New York. CHILDRESS,S, and SOWARD, M. (1972). Convection driven hydromagnetic dynamo. Phys. A. Rev. Lett. 29, 837-839. COWLING, G. (1934).The magnetic field of sunspots, Mon. Notic. Roy. Astron. Soc. 9 4 , 3 9 4 . T. DIENZER, W., VON KUSSEROW, U., and STIX, M. (1974). Steady and oscillatory am-dynamos. H. Astron. Astrophys. 36, 69-78. DILLON, F. (1975). Gravity and magnetic field correlation and related geomagnetic topics. R. Ph.D. Thesis, Cambridge Univ., Cambridge, England. DOLGINOV,. (1975). On Magnetic Dynamo Mechanism of the Planets, Prep. N9a (124). S Inst. Terr. Magn., Ionosphere, Radio Wave Propagation, Moscow. DOLGINOV, YEROSHENKO, and ZHUZGOV, (1973). Magnetic field in the very close S., Y., L. neighbourhood of Mars according to data from the Mars 2 and Mars 3 Spacecraft. J . Geophys. Res. 78, 4779-4786. DROBYSHEVSKI, M., and YUFEREV, S (1974). Topological pumping of magnetic flux by E. V. . three-dimensional convection. J. Fluid Mech. 65, 33-44. ELTAYEB, A. (1972). Hydromagnetic convection in a rapidly rotating fluid layer. Proc. Roy. I. Soc., Ser. A 326, 229-254. ELTAYEB, A. (1975). Overstable hydromagnetic convection in a rotating fluid layer. J. Fluid I. Mech. 71, 161-179. FURTH, P., KILLEEN, and ROSENBLUTH, N. (1963). Finite-resitivity instab H. J., M. sheet pinch. Phys. Fluids 6,459-484. GAILITIS, (1970).Selfexcitation of a magnetic field by a pair of annular vortices. Mag. Gidrod. A. 6, 19. [Magnetohydrodynamics 6, 14 (1972).] GIBSON, D. (1968).The Herzenberg dynamo 11. Quart. J. Mech. Appl. Math. 21, 257-267. R. GOLITSYN, S. (1960). Fluctuations of the magnetic field and current density in a turbulent G. flow of a weakly conducting fluid. Sou. Phys. Dokl. 5, 536-539.
118
H.K . Moffatt
a
>
I I
GUBBINS, (1974). Theories of the geomagnetic and solar dynamos. Rev. Geophys. Space Phys. D. 12, 137-154. HERZENBERG, (1958). Geomagnetic dynamos. Phil. Trans. Roy. Soc. London, Ser. A 250, A. 543-585. HIDE,R. (1974). Jupiter and Saturn. Proc. Roy. Soc., Ser. A 336, 63-84. HIDE,R., and ACHESON, J. (1973). Hydromagnetics of rotating fluids. Rep. Progr. Phys. 36, D. 159-221. HIDE, R., and ROBERTS, H. (1961). The origin of the main geomagnetic field. Phys. Chem. P. Earth 4, 25-98. HIGGINS,G. H., and KENNEDY, C. (1971). The adiabatic gradient and the melting point G. gradient in the core of the Earth. J. Geophys. Res. 76, 1870-1878. IBBETSON, and TRIITON, D. J. (1975). Experiments on turbulence in a rotating fluid. J. Fluid A., Mech. 68, 639-672. JEPPS,S. A. (1975). Numerical models of hydromagnetic dynamos. J. Fluid Mech. 67,625-646. KENNEDY, C., and HIGGINS, H. (1973). The core paradox, J. Geophys. Res. 7 8 , 9 W 9 0 4 . G. G. KIEPENHEUEQ 0.(1953). Solar activity. In The Sun (G. P. Kuiper, ed.), pp. 322-465. Univ. K. of Chicago Press, Chicago, Illinois. KOVASZNAY, S. G. (1960). Plasma turbulence. Rev. Mod. Phys. 32, 815-822. L. KRAICHNAN, H. (1973). Helical turbulence and absolute equilibrium. J. Fluid Mech. 59, R. 745-752. KRAICHNAN, H., and NAGARAJAN, (1967). Growth of turbulent magnetic fields. Phys. R. S. Fluids 10, 859-870. KRAUSF, F. (1968). Eine Losung des Dynamo problems auf der Grundlage einer linearen Theorie der magnetohydrodynamischen Turbulenz. Habilitationschrift, Univ. of Jena, Jena. (See Roberts and Stix, 1971.) KRAUSF,F., and ROBERTS, H. (1973). Some problems of mean field electrodynamics. P. Astrophys. J. 181, 977-992. KRAUSE, and STEENBECK, (1967). Some simple models of magnetic field regeneration by F., M. non-mirror-symmetric turbulence. Z. Naturforsch. A 22, 671-675. (See Roberts and Stix, 1971.) LEHNERT, (1954). Magnetohydrodynamic waves under the action of the Coriolis force. B. Astrophys. J. 119, 647-654. LEIGHTON, B. (1969). A magneto-kinematic model of the solar cycle. Astrophys. J. 156, R. 1-26. LERCHE, (1971a). Kinematic-dynamo theory. Astrophys. J. 166, 627-638. I. LERCHE, (1971b). Kinematic-dynamo theory, 11. Dynamo action in infinite media with isotroI. pic turbulence. Astrophvs. J. 166, 639-649. LIEPMANN, W. (1952). Askcts of the turbulence problem. Z. Angew. Math. Phys. 3,321-342. H. LIGHTHILL, J. (1959). Introduction to Fourier Transforms and Generalized Functions. M. Cambridge Univ. Press, London and New York. LORTZ, (1968). Impossibility of steady dynamos with certain symmetries. Phys. Fluids 11, D. 91 3-9 1 5. LOWES, J., and WILKINSON, (1963). Geomagnetic dynamo: A laboratory model. Nature F. I. (London) 198, 1158-1160. L o w s , F. J., and WILKINSON,(1968). Geomagnetic dynamo: An improved laboratory model. I. Nature (London) 219, 717-718. LUMLEY, L. (1962). The mathematical nature of the problem of relating Lagrangian and J. Eulerian statistical functions in turbulence. I n Mecanique de la Turbulence, No. 108, pp. 17-26. CNRS, Paris. MALKUS, V. R. (1959). Magnetoconvection in a viscous fluid of infinite electrical conductiW. vity. Astrophys. J. 130, 259-275.
v
J
t .
i.
9
1
179
MALKUS, V. R., and PROCTOR, R. E. (1975). The macrodynamics of a-effect dynamos in W. M. rotating fluids. J. Fluid Mech. 67, 417-444. MAUNDER, W. (1922). The Sun and sunspots 1820-1920. Mon. Not. Roy. Astron. Soc. 82, E. 5 34- 544. MOFFATT, K. (1961). The amplification of a weak applied magnetic field by turbulence in H. a fluids of moderate conductivity. J. Fluid Mech. 11, 625-635. MOFFATT, K. (1963). Magnetic eddies in an incompressible viscous fluid of high electrical H. conductivity. J . Fluid Mech. 17, 225-239. MOFFATT, K. (1969). The degree of knottedness of tangled vortex lines. J . Fluid Mech. 35, H. 117-129. MOFFATT, K. (1970a). Turbulent dynamo action at low magnetic Reynolds number. J. Fluid H. Mech. 41, 435-452. MOFFATT, K. (1970b). Dynamo action associated with random inertial waves in a rotating H. conducting fluid. J. Fluid Mech. 44, 705-719. MOFFATT, K. (1972). An approach to a dynamic theory of dynamo action in a rotating H. conducting fluid. J. Fluid Mech. 53, 385-399. MOFFATT, K., ed. (1973). Report on the NATO Advanced Study Institute on magnetohyH. drodynamic phenomena in rotating fluids. J. Fluid Mech. 57, 625-649. MOFFATT, K. (1974). The mean electromotive force generated by turbulence in the limit of H. perfect conductivity. J. Fluid Mech. 65, 1-10. NESS,N., BEHANNON, W., LEPPING,R. P., WHANG,Y. C., and SCHATTEN, H. (1974). K. K. Magnetic field observations near Mercury: Preliminary results from Mariner 10. Science 185, 151-160. PARKER, N. (195Sa). Hydromagnetic dynamo models. Astrophys. J. 122, 293-314. E. PARKER, N. (1955b). The formation of sunspots from the solar toroidal field. Astrophys. J. E. 121, 491-507. PARKER, N. (1963). Kinematical hydromagnetic theory and its applications to the low solar E. s photosphere. Astrophys. J. 138, 552-575. E. PARKER, N.(1970a).The origin of Solar magnetic fields. Ann. Rev. Astron. Astrophys. 8.1-30. PARKER, N. (1970b).The generation of magnetic fields in astrophysical bodies. I. The dynamo E. equations. Astrophys. J. 162, 665-673. PARKER, N. (1971a). The generation of magnetic fields in astrophysical bodies. 11. The E. galactic field. Astrophys. J. 163, 255-278. E. PARKER, N. (1971b). The generation of magnetic fields in astrophysical bodies. 111. Turbulent diffusion of fields and efficient dynamos. Astrophys. J. 163, 279-285. .PARKER, E. N. (1971~). generation of magnetic fields in astrophysical bodies. IV. The Solar The and terrestrial dynamos. Astrophys. J. 164, 491-509. E. PARKER, N. (1971d). The generation of magnetic fields in astrophysical bodies. V. Behaviour at large dynamo numbers. Astrophys. J. 165, 139-146. PARKER,N. (1971e). The generation of magnetic fields in astrophysical bodies. VI. Periodic E. modes of the galactic field. Astrophys. J. 166, 295-300. PARKER,N. (1971f). The generation of magnetic fields in astrophysical bodies. VIII. DynamiE. cal considerations Astrophys. J. 168, 239-249. PARKER, N. (1972). Topological dissipation and the small-scale fields in turbulent gases. E. # Astrophys. J. 174, 499-510. E. PARKER, N. (1973). The reconnection rate of magnetic fields. Astrophys. J. 180, 247. PARKER, N. (1975). Generation of magnetic fields in astrophysical bodies, X. Magnetic E. buoyancy and the solar dynamo. Astrophys. J. 198, 205-209. R. PARKER, L. (1966). Reconnexion of lines of force in rotating spheres and cylinders. Proc. Roy. Soc., Ser. A 291, 60-72.
8
180
H . K . MofSatt
PICHAKHCHI,D. (1966). Theory of the hydromagnetic dynamo. Sou. Phys.-JETP 23, L. 542-543. PIDDINGTON, H. (1972a). Solar dynamo theory and the models of Babcock and highton. Sol. J. Phys. 22, 3-19. PIDDINGTON, H. (1972b). The origin and form of the galactic magnetic field, I. Parkers J. dynamo model. Cosmic Electrodyn. 3, 60-70. PROCTOR, M. R. E. (1975). Non-linear mean field dynamo models and related topics. Ph.D. Thesis, Cambridge Univ., Cambridge, England. RADLER,K.-H. (1969a). On the electrodynamics of turbulent fluids under the influence of Coriolis forces. Monats. Dt. Akad. Wiss. Berlin 11, 194-201 (see Roberts and Stix, 1971). RADLER, K.-H. (1969b). A new turbulent dynamo, I. Monatsber. Deut. A h d . Wiss. Berlin 11, 272-279. (See Roberts and Stix, 1971.) ROBERTS, 0. (1970). Spatially periodic dynamos. Phil. Trans. Roy. Soc. London, Ser. A 266, G. 535-558. ROBERTS, 0.(1972). Dynamo action of fluid motions with two-dimensional periodicity. Phil. G. Trans. Roy. Soc. London, Ser. A 211, 41 1-454. ROBERTS, H. (1967). An Introduction to Magnetohydrodynamics. Longmans, Green, New P. York. ROBERTS, H. (1971). Dynamo theory. In Mathematical Problems in the Geophysical P. Sciences (W. H. Reid, ed.), pp. 129-206. Amer. Math. Soc., Providence Rhode Island. ROBERTS, H. (1972). Kinematic dynamo models. Phil. Trans. Roy. Soc. London, Ser. A 272, P. 663-698. ROBERTS, H., and SOWARD, M. (1972). Magnetohydrodynamics of the Earths core. Annu. P. A. Rev. Fluid Mech. 4, 117-154. ROBERTS, H., and SOWARD, M. (1975). A unified approach to mean field electrodynamics. P. A. Astron. Nachr. 2%, 49-64. P. K. ROBERTS, H., and STEWARTSON, (1974). On finite amplitude convection in a rotating magnetic system. Phil. Trans. Roy. Soc. London, Ser. A 211, 287-315. K. ROBERTS, H., and STEWARTSON, (1975). On double roll convection in a rotating magnetic P. system. J. Fluid Mech. 68, 447-466. ROBERTS,P. H, and STIX, M. (1971). The Turbulent Dynamo: A Translation of a Series of Papers by F. Krause, K.-H. Radler and M. Steenbeck, Tech. Note NCAR/TN/lAdO. Nat. Cent. Atmos. Phys., Boulder, Colorado. P. ROBERTS, H., and STIX, M. (1972). a-Effect dynamos by the Bullard-Gellman formalism. Astron. Astrophys. 18, 453-466. SAFFMAN, G. (1960). On the effect of the molecular diffusivity in turbulent diffusion. J. Fluid P. Mech. 8, 273-283. SAFFMAN, G. (1963). On the fine-scale structure ofvector fields convected by a turbulent fluid. P. J . Fluid Mech. 16, 545-572. S C H L ~ E R , and BIERMANN, (1950). Interstellare Magnetfelder. 2. Naturforsch. A 5, A., L. 237-251. SMITH, DAVIS, JONES, E., COLBURN, S., E., L., D. D. COLEMAN, J., DYAL, and SONETT, P. P. P., C. (1974). Magnetic field of Jupiter and its interaction with the Solar wind. Science 183, 305-306. SOWARD, M. (1971). Nearly symmetric advection. J. Math. Phys. (N.Y.) 12, 2052-2062. A. SOWARD, M. (1972). A kinematic theory of large magnetic Reynolds number dynamos. Phil. A. Trans. Roy. Soc. London, Ser. A 272, 431-462. SOWARD, M. (1974). A convection-driven dynamo I. The weak field case. Phil. Trans. Roy. A. Soc. London, Ser. A 275, 611-651.
a
b
J
t
3
I
181
,.
A
SOWARD, M. (1975). Random waves and dynamo action. J. Fluid Mech. 69, 145-177. A. STEENBECK, and KRAUSE, (1966). The generation of stellar and planetary magnetic fields M., F. by turbulent dynamo action. Z. Nuturjorsch. A. 21,1285-1296. (See Roberts and Stix, 1971.) STEENBECK, and KRAUSE, (1969a). On the dynamo theory of stellar and planetary magM., F. netic fields, I. A.C. dynamos of solar type. Astron. Nachr. 291,49-84. (See Roberts and Stix, 1971.) STEENBECK, and KRAUSE, (1969b). On the dynamo theory of stellar and planetary M., F. magnetic fields, 11. D.C. dynamos of planetary type. Astron. Nuchr. 291, 271-286. (See Roberts and Stix, 1971.) F., K.-H. (1966). A calculation of the mean electromotive STEENBECK, KRAUSE, and R ~ L E R , M., force in an electrically conducting fluid in turbulent motion, under the influence of Coriolis forces. Z. Nuturforsch. A 21, 369-376. (See Roberts and Stix, 1971.) AYLOR,G. I. (1922). Diffusion by continuous movements. Proc. London Math. Soc., A 20, 196-211. (Collect. pap., Vol. 11, pp. 172-184. Cambridge Univ. Press, London and New York, 1960.) TAYLOR, B. (1963). The magnetohydrodynamics of a rotating fluid and the Earths dynamo J. problem. Proc. Roy. Soc. London, Ser. A 274, 274-283. TOUGH, G. (1967). Nearly symmetric dynamos. Geophys. J. Roy. Astron. Soc. 13, 393-396. J. TOUGH,J. G., and GIBSON, R. D. (1969). The Braginskii dynamo. I n The Application of Modern Physics to the Earth and Planetary Interiors (S. K. Runcorn, ed.), pp. 555-569. Wiley, New York. VAINSHTEIN, I., and ZELDOVICH, B. (1972). Origin of magnetic fields in astrophysics. Sou. S. Y. Phys.-Usp. 15, 159-172. VENEZIAN, (1967). Magnetohydrodynamics of the Earths Magnetic Field, Rep. No. 85-37, G. Div. Eng. Appl. Sci., California Inst. Tech., Pasadena, California. WARWICK,(1963). Dynamic spectra ofJupiters decametric emission. Astrophys. J. 137,41-60. J. WEISS,N. 0. (1966). The expulsion of magnetic flux by eddies. Proc. Roy. Soc., Ser. A 293, 310-328. WEISS,N. 0. (1971). The dynamo problem. Quart. J. Roy. Astron. Soc. 12, 432-446. WEISS, 0. (1974). Dynamo maintenance of magnetic fields in stars. I n MagnetohydrodynaN. mics (L. Mestel, F. Meyer, and N. 0.Weiss, eds.). Swiss Soc. Astron. Astrophys., Geneva. WOLTJER, (1958). A theorem on force-free magnetic fields. Proc. Nut. Acad. Sci. U S . 44, L. 489-49 1. ZELWVICH, B. (1957). The magnetic field in the two-dimensional motion of a conducting Y. turbulent liquid. Sou. Phys.-JETP 4, 460-462.

H.K. Moffatt - Generation of Magnetic Fields by Fluid Motion

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

H.K. Moffatt - Generation of Magnetic Fields by Fluid Motion

Uploaded by

Copyright:

Available Formats

www.moffatt.

ACADEMIC P&=. New York San Francisco

Generation of Magnetic Fields by Fluid Motion

Generation of Magnetic Fields by Fluid Motion

Dipole moment p (G km3)

Rotation period (days)

6371 71,351 3386 2439

* Dolginov et al. (1973).

Warwick (1963); Smith et al. (1974).

Generation of Magnetic Fields by Fluid Motion

Generation of Magnetic Fields by Fluid Motion

11. Magnetokinematic Preliminaries

A. IDEALIZATION KINEMATIC OF THE DYNAMO PROBLEM

(2.6) aB/at = - V A E , where po = 4n x 10- S.I. units. In p, where j = 0, B is determined by

Generation of Magnetic Fields by Fluid Motion

B. MAGNETIC FIELD REPRESENTATIONS

(2.1 1) (2.12) (2.13)

H. K. Mofatt 3. Two-Dimensional Fields

Generation of Magnetic Fields by Fluid Motion

D. NATURAL DECAY MODES AND FORCE-FREE FIELDS

SY + iSy = V A V A [ ~ ~ - ~ ' ~ J , + ~ , ~ ( A - ' ~ , ~ ~ ) ~ (2.21) ~~],

S(r, e) = Ar-2J3i2(Kr) COS 8,

S = - Bo(r - R3/r2) cos 8,

the latter condition ensuring smoothness across r = R.

111. Convection, Distortion, and Diffusion of B Lines

Generation of Magnetic Fields by Fluid Motion

A. BALANCE STRETCHING DIFFUSION A OF AND IN MAGNETIC FLUXROPE

The steady solution of (2.9) is then

FLUX EXPULSION FLOWS BY WITH

Generation o Magnetic Fields by Fluid Motion f

c. TOPOLOGICAL PUMPING OF MAGNETIC FLUX

+ f cos x) sin y cos z, (cos x + cos y + cos x cos y ) sin z),

Generation of Magnetic Fields by Fluid Motion

B4 (2C sin 6/31r2)

IV. Some Basic Results

A. COWLINGS THEOREM RELATED AND RESULTS

axisymmetric conditions [analogous to (3.35)] is

Generation of Magnetic. Fields by Fluid Motion

D(x * B)/Dt = (B V)(X * U)

Generation o Magnetic Fields by Fluid Motion f

V. The Mean Electromotive Force Generated by a Random Velocity Field

scale L of B , is very large, such a relationship may presumably be developed as a series

(5.5) field, a and /? are

Generation of Magnetic Fields by Fluid Motion

where Jo = (j), E, = (E), or equivalently,

Clearly we have exponential growth (i.e., dynamo action) if

B. THESTRONG DIFFUSION LIMIT

which may be neglected to lowest order. The resulting equation

ab/at = V A (U AB,) + lV2b

The spectrum tensor Qij(k) is related to p(k, t) by

Tij(k) = (lk2)-(B0 * k)2@ij(k).

Generation of Magnetic Fields by Fluid Motion

0 The Schwarz inequality

may be translated into spectral terms to show that, for all k,

aij = a$) - eijk5 .

FIG.3. Possible distortion of a line of force by a cyclonic event.

Generation of Magnetic Fields by Fluid Motion

C. EVALUATION aij AND OF

Bijk FOR A RANDOM WAVE FIELD

ii(k, w)ei(.x - O t ) d3k d o ,

to the following expressions for aij and &k:

dk(w2 + 12k4)-'k2F(k, w),

dk(w2+ L2k4)- 'k2E(k, w )